Maybe.

This old playbook worked 8% of the time.

It gave us antibiotics, vaccines, statins, antibodies, GLP-1s, cancer immunotherapies, antivirals, and rare-disease drugs. Modern medicine is a miracle. But the machine that creates it is slow, expensive, fragmented, and full of handoffs.

This is not because pharma is dumb. It is because pharma is hard. Biology is hard. Patients are different from each other. Clinical trials are expensive. Manufacturing is regulated. Quality matters. Regulators are cautious because they should be. The body is not software. You cannot ship a broken update into a human being.

But the company structure is also broken. Pharma is not one business. It is ten businesses wearing a lab coat. A pharma company has to discover drugs, test safety, design trials, recruit patients, collect data, manufacture medicine, handle quality, work with regulators, get paid by insurers, educate doctors, and monitor safety after launch. Historically, all of this has been split across vendors, consultants, spreadsheets, databases, hospitals, contract manufacturers, and internal teams.

Every handoff adds time. Every handoff adds cost. Every handoff adds error. Every handoff loses information.

The next great pharma company will not be a better version of this old machine. It will be a new machine: an AI-native pharma company.

machines that make medicine

machines that make medicine

Panoptic Bio is building a vertically integrated AI pharma company for the new era. Agent-run clinical trials, connected directly to automated biomanufacturing…

Read more

17 days ago · Sebastian De Beurs

By AI, I mean software that can read, reason, predict, draft, search, and help make decisions. By AI-native, I do not mean adding a chatbot to an old pharma workflow. I mean building the whole company around artificial intelligence from the beginning. AI proposes. Software executes. Robots run experiments. Simulations prioritize. Humans approve the high-risk decisions. Regulators receive a clear record of what happened and why.

That is the new playbook.

The first thing you build is not a molecule. The first thing you build is memory.

Memory
A pharma company is only as good as what it remembers: every paper, every failed trial, every assay, every patient group, every biomarker, every dose change, every protocol amendment, every safety issue, every manufacturing batch, every payer objection, every label update. Traditional pharma stores this memory in people, PDFs, email threads, folders, outsourced vendors, and institutional lore. The AI-native pharma company stores it in a scientific memory: one connected system that learns from every molecule, trial, patient, batch, launch, and safety signal.

This scientific memory becomes the company’s brain. The discovery system uses it to find drug ideas. The safety system uses it to predict risk. The trial system uses it to decide whether a drug is likely to work. The enrollment system uses it to find patients. The manufacturing system uses it to make the drug. The launch system uses it to understand the market. The quality system watches everything. This is the basic architecture of the vertically integrated AI-native pharma company: one memory, many specialized agents, and humans approving the decisions where safety and accountability matter.

But you do not build the whole thing on day one. That is how you die.

The smallest credible version starts with clinical trial intelligence. Clinical trial intelligence means understanding which drugs are likely to succeed in human trials and why.

This is the best place to start because the data already exists. There are public trial registries, medical papers, drug labels, patents, press releases, conference abstracts, regulatory documents, and years of failed and successful trials.

You do not need a wet lab to begin. You do not need a factory to begin. You do not need 500 people to begin. You need a system that can answer one question better than everyone else: will this drug work?

Not “is the science interesting?” Not “does the pitch deck look good?”

Will this drug work in this disease, in this patient population, at this dose, against this comparator, with this endpoint, under this regulatory path?

This is where the company begins. Not with a molecule. With judgment.

The first product should feel like Bloomberg Terminal for drugs. Panoptic Bio calls it the Trial Terminal.

A clinical asset goes in, and a decision comes out. The input is a drug, target, disease, trial, or company. The output is a complete map: what the drug does, what failed before, what worked before, which patients matter, which endpoints are credible, which competitors are ahead, which trial design makes sense, what the risks are, and whether the asset is worth buying, licensing, redesigning, or killing.

A human analyst can do this slowly. A great team can do this expensively. An AI-native pharma company should do this continuously.

AI drug rescue.

The first AI-native pharma companies do not need to invent every drug from scratch. They can buy mistakes. Pharma creates mistakes constantly. Good molecules are tested in the wrong disease. Good mechanisms are tested in the wrong patients. Good assets are killed after underpowered trials. Good drugs fail because the endpoint was wrong. Good programs get stuck because manufacturing was messy or enrollment was too slow.

The opportunity is to find the asset everyone else thinks is dead and ask why it died. Was it biology? Was it dose? Was it endpoint? Was it patient selection? Was it manufacturing? Was it commercial? Or was it simply owned by the wrong company at the wrong time?

Then you produce the only memo that matters: here is the asset, here is why it failed, here is why we think it can work, here is the redesigned trial, and here is the probability that we are wrong.

That is a pharma company.

Protocols
Once you know the asset, you need the trial. This is where old pharma becomes obviously insane. A clinical protocol is one of the most important documents in medicine. A protocol is the master plan for a clinical trial. It defines who can join, what dose they get, what doctors measure, how safety is monitored, how often patients visit, what counts as success, and what regulators will review.

And yet protocols are often written like legal documents passed through a committee of 47 people who each add risk but not clarity.

The AI-native pharma company builds a Protocol Machine. The Protocol Machine turns an asset strategy into a trial plan. It studies similar successful and failed trials. It asks which patient criteria slowed enrollment, which endpoints regulators accepted, which amendments happened later and should have been obvious earlier, and which design choices created unnecessary burden for patients.

The AI system writes the first draft. The statistician checks the math. The physician checks the medicine. The regulatory lead checks the rules. The operations team checks whether the trial can actually run. Then humans approve it.

That pattern matters: AI drafts, AI simulates, AI stress-tests, humans approve.

AI-native does not mean AI-only. Especially not in medicine. The future is not autonomous agents secretly dosing patients and releasing drugs. That is science fiction and also a nightmare. The future is a company where AI compresses the work, expands the search space, catches mistakes, keeps the evidence trail, and makes expert humans dramatically more effective.

Enrollment
The next task is enrollment. A trial without patients is not a trial. It is a PDF.

Clinical trials often fail slowly because they cannot find enough eligible patients at the right sites at the right time. The old way is site surveys, investigator relationships, optimistic spreadsheets, manual chart review, and prayer.

The AI-native way is patient-trial matching. The trial criteria become computable. The system looks across approved medical data sources and identifies which hospitals are likely to have eligible patients. It estimates screen failure, meaning how many patients appear eligible at first but fail once fully reviewed. It flags competing trials. It predicts enrollment speed. It helps coordinators find likely matches.

But it does not bypass ethics. Patients are not ad targets. A site investigator and coordinator still approve contact. Patients still consent. Doctors still make the final eligibility decision.

The company that controls enrollment intelligence controls clinical speed. And speed in pharma is life.

Making the medicine
After digital AI comes physical AI. Digital AI is software that reasons. Physical AI is software connected to robots, sensors, lab equipment, and manufacturing systems. This is where the next pharma company becomes very different from a software company.

Software can find the drug. Software can design the trial. Software can match the patient. But at some point, atoms matter. You have to make the medicine at clinical grade, under regulation, reliably, repeatedly, and at scale.

This is where the current system breaks again. Clinical development and manufacturing are treated like separate worlds. They are not. They are one loop.

The trial tells you what dose you need. The dose tells you how much drug you need. The formulation affects whether patients take it correctly. The manufacturing process affects purity. Purity affects safety. Safety affects dose. Dose affects efficacy. Efficacy affects trial design. Trial design affects manufacturing demand.

Manufacturing is not downstream of clinical development. Manufacturing is part of clinical development.

Eventually, the AI-native pharma company needs a BioFactory OS. OS means operating system: the control layer that coordinates the work. A BioFactory OS connects process development, sensors, batch records, quality checks, release testing, supply planning, and automated manufacturing. It watches the process in real time. It predicts when a batch may fail. It recommends adjustments. It flags deviations. It prepares quality review.

But it never releases medicine without human approval. Quality assurance still matters. Regulators still matter. Human accountability still matters.

This is the bridge from digital AI to physical AI. Machines do not just discover medicine. Machines make medicine. Clinical trials and manufacturing are not separate systems. They are one loop, and the next company should connect them directly.

Taking medicine to patients!
Most biotech founders think the story ends at approval. It does not. Approval is when the second company begins.

Now you need market access, meaning the work required to get insurers and health systems to pay for the drug. You need medical affairs, meaning the team that communicates scientific evidence to doctors and researchers. You need pharmacovigilance, meaning post-market safety monitoring. You need supply, distribution, adherence, real-world evidence, and label expansion.

Real-world evidence means evidence from actual clinical use after approval: claims, prescriptions, lab results, registries, doctor notes, and patient outcomes. The AI-native pharma company treats the market as the next learning loop. Every prescription, claim, safety report, refill, discontinuation, physician question, payer objection, and patient outcome teaches the system something.

Traditional pharma runs a trial, launches a drug, and then collects post-market data because it has to. AI-native pharma launches the drug and keeps learning because that is the whole point.

So how do you actually launch?

First, build the Trial Intelligence and Due Diligence Engine. Use public and licensed data. Analyze active and failed trials in one therapeutic area. Predict probability of success. Diagnose failure modes. Write asset memos. Find mispriced drugs.

Second, acquire or license one drug. Not ten. One. The best first asset is not the flashiest asset. It is the one where the failure mode is fixable: wrong endpoint, wrong patient population, wrong dose, poor enrollment, bad trial design, or a manufacturing issue you can solve.

Third, redesign the trial. Make the protocol computable from the start. Use AI to draft, simulate, and stress-test. Use humans to approve. Meet regulators early. Design the trial so the machine can help run it.

Fourth, build enrollment infrastructure. Partner with sites. Map patients compliantly. Reduce screen failure. Track bottlenecks in real time. Make enrollment an operating system, not a weekly meeting.

Fifth, connect manufacturing. At first, partner with manufacturers. Then automate process development for one type of medicine. Do not build a factory for everything. Pick one modality, meaning one category of medicine, such as antibodies, small molecules, recombinant proteins, mRNA medicines, viral vectors, or cell therapies. Instrument everything. Make every batch teach the system.

Sixth, launch with real-world evidence from day one. Market access is not a deck. Medical affairs is not a call center. Safety monitoring is not paperwork. They are all part of the learning loop.

The first team does not need to look like a classic pharma org chart. You need a small team that can cross software, biology, medicine, regulation, and operations. You need a founder who can raise money and hold the whole system in their head. You need a clinical development lead, a physician-scientist, a regulatory lead, a machine-learning lead, a data-infrastructure lead, a product engineer, a business-development lead, a manufacturing advisor, and a quality-minded operator who understands that pharma is not SaaS.

A lot of companies want to sell SaaS tools to pharma. That is fine. But the bigger opportunity is not to sell dashboards to the old system. The bigger opportunity is to become the new system.

machines that make medicine
The next great pharma company will look like a self-driving machine. It will have a brain that reads every paper, trial, label, protocol, patent, safety report, and batch record. It will have AI agents that reason across discovery, safety, clinical development, enrollment, manufacturing, regulation, and launch. It will have robots and sensors that make biology measurable and manufacturing repeatable. It will have humans at the gates where judgment, ethics, safety, and accountability matter.

The old pharma company asks: can we discover a molecule?

The new pharma company asks: can we build a machine that continuously discovers, tests, manufactures, launches, and learns from medicine?

That is how you launch your own pharma company. Not by starting with a drug, but by starting with the loop.

Hypothesis → AI design → simulation → robotic experiment → evidence update → human approval → trial → manufacturing → launch → real-world learning.

Then repeat.

Here’s a peak at how we do it at Panoptic Bio:

Expensive. Slow. And failure was the norm. The industry had optimized itself around massive programs, complex handoffs, long development cycles, and no room for fast iteration.

SpaceX was founded in 2002. Not as another supplier, better consultant, or marginally cheaper contractor. Instead, SpaceX rebuilt the launch system itself around success.

It questioned the requirements. It deleted unnecessary complexity. SpaceX’s breakthrough was making rockets reusable and repeatable. Launch, land, inspect, and relaunch. Turning spaceflight from a one-time event into a repeatable system that dramatically improved success, speed, and cost. SpaceX learned by launching, failing, relaunching, and improving. And over time, the company turned one of the most expensive, slow-moving engineering domains in the world into something repeatable.

The equivalent breakthrough in pharma would be turning drug development from isolated, mostly failed experiments into a repeatable learning system where every trial improves the odds that the next drug successfully reaches patients.

Pharma in 2026 is much like Aerospace in 2002.

Eli Lilly is the Boeing of today. The market leader, with a market cap of roughly $1T. Founded in 1876, with over 50,000 employees, it is a legacy player.

The world is spending more than ever on pharmaceuticals and pharmaceutical R&D. The industry is enormous. The science extraordinary. And the need is obvious.

And yet, the system underneath is deeply inefficient.

The industry often behaves as if the main problem is in drug discovery. Discovery gets most of the attention because it is exciting. It is visual. It is computational. It is where foundation models, generative chemistry, target discovery, and biological data platforms naturally cluster.

The brutal reality is that only a small percentage of clinical trials bring a new treatment to market.

The industry is filled with brilliant scientists, clinicians, operators, investors, and executives. But the decision-making infrastructure around clinical development remains fragmented.

Biology lives in one place. Clinical trial data lives somewhere else. Regulatory precedent is scattered. Commercial context is often downstream. Competitive intelligence is manually assembled. Financial assumptions are rebuilt in spreadsheets. Analyst judgment lives in decks, notes, calls, and memory.

The result is that the most important decisions in the industry are still made through a messy combination of expert intuition, incomplete evidence, manual synthesis, and institutional pattern recognition. That is not good enough for a trillion-dollar industry. Especially not an industry where the cost of being wrong is measured in years, billions of dollars, and hundreds of thousands of patients waiting for better treatments.

The wrong optimization

For years, pharma has focused on speed and cost. Faster discovery. Cheaper trials. More automation. Resulting in more molecules, targets, screens, dashboards, and summaries. All of that matters. But it misses the central problem: You cannot optimize a system that is not successful.

Elon Musk’s engineering framework is simple:

The order matters. Acceleration and automation are steps 4 and 5; they come last.

Pharma has mostly done the opposite. Starting with automation. AI is being layered on top of broken workflows. More software is being added to processes that may not deserve to exist. More dashboards are being built for decisions that still lack the right underlying intelligence. That is why so many AI tools in pharma feel useful but not transformative. They make the existing system easier to operate.

But SpaceX did not win by making PowerPoints for Boeing. It won by rebuilding the launch system. That is what pharma needs. Not another tool around the edges. Not another workflow product. Not another chatbot on fragmented data.

Pharma needs a repeatable launch and relaunch system for drug development. Right now, the industry has an 8% success rate. It is not ripe for automation. It is ripe for questioning constraints, deleting what should not exist, and optimizing for success rate. A faster failed trial is still a failed trial. A cheaper failed trial is still a failed trial. An automated failed trial is still a failed trial. The glaring error is success rate. And therefore, the glaring opportunity is success rate. The future is in clinical.

The SpaceX of pharma

The lesson of SpaceX is not “move fast.” That is the shallow reading.

The deeper lesson is that old industries are not transformed by optimizing individual pieces of a broken system. They are transformed when someone rebuilds the system around a better operating model. SpaceX did not merely make rockets cheaper. It created a launch and relaunch machine.

Pharma needs the same thing: A repeat-success machine.

This is what the pharma company of the future looks like:

The floors are arranged in the order of the optimal drug development lifecycle for the AI era. The top floor of this pharma company is the brain. A decision making AI that understands risk, evidence, biology, regulation, competition, and value. It is a system where every trial makes the next trial smarter. A system where every failure teaches the model. A system where evidence compounds.The next floor is a data and computation center, because data and compute are the new assets of pharma, not just bench scientists. The next floor is a discovery lab, where fundamental and translational science occurs. The next floor is for in-human clinical testing. The next floor is for biomanufacturing. And the ground floor is for shipping. This pharma company is all connected. All learning. All oriented around one goal: increasing the probability that important medicines actually reach patients.

This year is the year the SpaceX of pharma is being founded

It will not look like a traditional pharma company at first. It may not even look like a pharma company at all. It will start as infrastructure. It will start as software. It will start as a terminal for better decisions.

Before the rocket lands, there is a control room. That is how these transformations often begin. Before the company changes the industry, it changes how the industry makes decisions. Pharma does not need another chatbot. It needs its SpaceX. A company that can launch and re-launch drugs to patients like rockets to the moon.

Panoptic Bio team member showing results of the top ranking benchmark in clinical trial outcome prediction.

Pharma Disruption starts in 2026

The old aerospace industry was a maze of suppliers, contractors, consultants, and middlemen. Everyone got paid. Not everyone learned. Pharma is fragmented the same way. The winner won’t be the company that outsources development most efficiently. It will be the company that captures and compounds every lesson from every trial.

History is usually changed by small teams willing to take on impossibly large problems.

Sources:

Clinical Development Success Rates and Contributing Factors 2011-2020
https://go.bio.org/rs/490-EHZ-999/images/ClinicalDevelopmentSuccessRates2011_2020.pdf

Dynamic clinical trial success rates for drugs in the 21st century
https://www.nature.com/articles/s41467-025-64552-2

Global R&D Trends 2026 https://www.iqvia.com/insights/the-iqvia-institute/reports-and-publications/reports/global-r-and-d-trends-2026

Measuring the return from pharmaceutical innovation 2024 https://www.deloitte.com/ch/en/Industries/life-sciences-health-care/analysis/measuring-the-return-from-pharmaceutical-innovation.html

Making more medicines that matter: A new recipe for biopharmaceutical R&D https://www.mckinsey.com/industries/life-sciences/our-insights/making-more-medicines-that-matter

Innovation in the pharmaceutical industry: New estimates of R&D costs
https://www.sciencedirect.com/science/article/abs/pii/S0167629616000291

Artificial intelligence in drug development
https://www.nature.com/articles/s41591-024-03434-4

How successful are AI-discovered drugs in clinical trials? A first analysis and emerging lessons
https://www.sciencedirect.com/science/article/pii/S135964462400134X

Digital in R&D: The $100 billion opportunity
https://www.mckinsey.de/~/media/McKinsey/Industries/Pharmaceuticals%20and%20Medical%20Products/Our%20Insights/Digital%20in%20R%20and%20D%20The%20100%20billion%20opportunity/Digital-in-R-and-D-The-100-billion-opportunity.pdf

Watch SpaceX’s reused Falcon 9 rocket nail its landing
https://www.wired.com/story/spacex-launch-watch-reuse-rocket/

www.panoptic.bio

We can generate data at scale. We can sequence almost anything. We can run trials across the world. We can train models on vast corpora of papers, patents, and protocols.

Still, most drugs fail.

Not because we lack information. Because we fail to turn information into judgment.

That is the real problem in drug development. Not data scarcity. Not compute. Judgment.

Biomedical intelligence, whether human or machine, lives or dies on this point. It must work across a stack of decisions that determine whether a therapy succeeds or fails. That stack is not vague. It has shape. It can be named.

It has twelve dimensions.

Each one captures a different layer of reality in life sciences. Together they decide whether a medicine works, gets approved, reaches patients, and creates value. Clinical trial intelligence will not matter until it can handle all twelve. Not one by one. Together.

1. Mechanism of action

This is the base layer. If you do not understand why a drug should work, the rest is built on sand.

A serious system has to start with the facts. What is the trial. Who is the sponsor. What is the intervention. What is the modality. What is the dose. What is the molecular target. What disease is being treated.

But that is just the beginning.

The real task is to move from labels to cause. Not summary. Cause.

You need the human biology of the target. Genetics. Expression. Disease linkage. You need pathway biology. Is the target upstream or downstream. Is it central or peripheral. You need target validation. Knockout. Knockdown. Pharmacologic modulation. You need biomarker evidence in humans that shows the drug is hitting what it claims to hit. And you need history. What else has gone after this target or pathway. What worked. What failed. What failed more than once.

The point is not to gather papers. It is to build a causal case.

This is hard because mechanism is never one question. It is a set of questions. What role does the target actually play in disease. How strong is the human evidence. Is the target a driver, or only a correlate. Are the findings consistent across genetics, cell work, animal work, and pharmacology. Where does the target sit in the pathway. Does that make it powerful, or fragile.

An upstream target may offer more leverage. It may also carry more risk. A downstream target may be safer. It may also matter less.

Then comes translation. Do biomarkers show target engagement in humans. Is there a PK PD relationship. Does biology move with dose.

Then comes class history. Has this kind of target worked before. Has it failed before. Has it failed for the same reason, again and again.

This is why mechanism cannot be a literature review. It has to be a judgment. You are sorting biology into levels of conviction. Well validated. Plausible but unproven. Weak. Speculative. That is the real job.

2. Efficacy

This is where biology meets the clinic.

Once you leave mechanism, the standard gets harder. A good story is no longer enough. Now the question is simple. Did the drug work in humans.

Not in a headline. Not in a press release. In the data.

A real system has to pull the whole record. Phase 1. Phase 2. Phase 3. Posters. Publications. Company updates. Failed studies too. Especially failed studies. It also has to look at similar drugs in the same class, because no efficacy result means much in a vacuum.

The question is not whether the trial sounded positive. The question is whether the signal survives contact with scrutiny.

That means reading the trial like an operator would. What phase was it. Was it randomized. What was the control arm. Was the study powered well. How long was follow up. What were the primary, secondary, and exploratory endpoints.

Then comes interpretation. Was the primary endpoint met. Were the p values and confidence intervals solid. Was multiplicity handled well. Did the result hold up across sensitivity analyses. Was there subgroup consistency, or did the signal depend on slicing the data until something looked good.

Then comes the harder question. Was the effect meaningful. Not statistically nonrandom. Meaningful.

A small effect can be real and still not matter. A large effect can matter even if the dataset is imperfect. This is where weak systems fail. They confuse significance with importance.

You also have to benchmark against standard of care. What do approved drugs achieve in this indication. What endpoints did they win on. What effect sizes were enough to change practice. Without that comparison, you mistake motion for progress.

Efficacy also has a time axis. A fragile signal often looks good once. Then breaks. So you have to ask whether the evidence is consistent across studies, across doses, across populations. Does the clinical effect match the mechanism, or is there a gap between preclinical promise and human reality. Are there signs of fragility. Small N. Post hoc wins. Endpoint switching. Regression to the mean.

Efficacy is not a binary variable. It is a judgment about whether a signal is durable enough to survive the next trial, the next population, and the next round of scrutiny. That is what matters.

3. Safety

Every drug has a cost.

The question is whether the cost is worth paying.

Safety is where many programs break, often quietly. A drug can hit the biology. It can even show efficacy. But if the exposure needed to work is too toxic, the program dies.

That is why safety cannot be treated as a side note. It is its own dimension.

You start with the exposure base. What trial is this. What drug. What dose. What schedule. What route. What phase. How many patients got each dose. For how long. Was the safety evidence generated in healthy volunteers, patients, or both.

Then you build the record. Treatment emergent adverse events. Grade 3 and 4 events. Serious adverse events. Discontinuations. Dose interruptions. Dose reductions. Withdrawals. Deaths. Prior studies matter. Failed studies matter. Related drugs matter. Regulatory warnings matter. Preclinical tox matters when it helps explain clinical risk.

But the real work starts after collection.

How good is the safety database. Is the patient exposure deep enough to say anything with confidence. Is there a maximum tolerated dose. Were there dose limiting toxicities. Is there a dose toxicity relationship. Is the efficacious dose inside a workable window, or beyond it.

Then comes interpretation. Are the toxicities on target or off target. Expected or surprising. Reversible or cumulative. Easy to monitor or hard to catch. Manageable in practice, or only manageable on paper.

You also have to think by organ system. Liver. Heart. QT. CNS. Blood. Immune system. GI. Kidney. Eye. Infection. Then modality specific risks. Gene therapy. Cell therapy. Radiopharma. ADCs. Oligos. Immune activation. Each class has its own ways to fail.

And safety is never absolute. It is always contextual.

A toxicity that is acceptable in late stage oncology may be disqualifying in migraine. A burden that physicians tolerate in a fatal disease may be unacceptable in a chronic one. So you have to benchmark against placebo, active control, and standard of care. Better. Similar. Worse. Acceptable for the disease. Or not.

Then comes the downstream effect. Could the profile trigger a clinical hold. A boxed warning. A REMS. A narrow label. Heavy monitoring. Commercial friction. Could a subgroup risk shrink the usable population.

Safety is not event counting. It is judgment under tradeoff. What risk. At what dose. In what disease. For what benefit. Against what alternatives. That is the real question.

4. Protocol design

A trial is not just an experiment.

It is a claim about what will count as truth.

If the design is weak, the result can be useless even if the drug works. That is why protocol design matters so much. It is where intent becomes structure.

You have to reconstruct the whole trial. Phase. Study design. Intervention. Endpoints. Sample size. Treatment duration. Inclusion and exclusion criteria. That is not clerical work. That is the skeleton of the decision the trial is trying to make.

Then you ask what kind of trial this really is. Signal finding. Dose finding. Proof of concept. Registrational. Supportive. These are not the same. A fuzzy Phase 2 study can still be useful. A fuzzy registrational study is a disaster.

Endpoint choice is everything. Is the primary endpoint clinically meaningful. Is it accepted by regulators. Is it a surrogate that may or may not translate. Has it supported approval before. Are the secondary endpoints aligned, or just noise around the edges. Timing matters too. Measure too early and you miss the effect. Too late and noise creeps in.

Control arm choice is where many weak programs hide. Placebo versus active control is not a minor detail. It shapes whether the result means anything. Randomization matters. Blinding matters. Bias matters.

Population selection matters just as much. A trial can be enriched for signal. That is sometimes smart. But it can also be so optimized that it stops reflecting the real patients who would get the drug later. Then the program wins early and breaks later.

Statistical design decides how fragile the result will be. Sample size. Power assumptions. Effect size assumptions. Multiplicity control. Interim looks. Missing data. Rescue meds. Dropouts. A study can be nominally positive and still not survive real scrutiny if these parts are weak.

And then there is the part people forget. Operations. Visit burden. Biomarker testing. Imaging. Procedures. Complexity at sites. Complexity for patients. The more burden you add, the more ways the data can fail.

In the end, protocol design has one real test. If this trial is positive, will anyone believe it. And will it be enough to move the drug forward.

5. Enrollment

This is where theory meets the world.

It is easy to say a disease has many patients. It is much harder to recruit them.

Enrollment is where many programs lose years. Sometimes they lose the whole program. Not because the science failed. Because the trial could not fill.

You start with scope. How many patients are needed. How many sites. Where are they. What is the timeline. Is the study recruiting. Stalled. Delayed. This tells you how hard the task really is.

Then you turn epidemiology into a funnel. Start with prevalence. Then cut it down. Diagnosed patients. Treated patients. Eligible patients. Biomarker positive patients. Patients willing to join. What looks large on paper often becomes small in practice.

Incidence versus prevalence matters too. Acute diseases recruit differently from chronic ones. Geography matters. Patients may exist, but not near the sites that matter.

Eligibility is often the hidden trap. Tight criteria. Prior line requirements. Biomarker screens. Complex procedures. All of these drive screen failures. Then comes competition. Other trials may be recruiting the same patients from the same sites at the same time. Approved therapies may reduce willingness to randomize. Good sites are finite. Overload them and everything slows.

Site strategy matters. Number of sites. Quality of sites. Academic versus community. Geographic spread. Activation speed. Dependence on a few star centers. These are not side issues. They decide whether the timeline is real.

Benchmarks matter most. Patients per site per month is real. It varies by disease and geography. If your plan depends on enrollment rates above what history supports, your plan is weak, no matter how pretty the slide looks.

And enrollment is never just about speed. Delay changes everything. It burns cash. It shifts readouts. It can turn a first mover into a follower. It can move a promising asset from relevant to late.

Many drugs do not fail on science. They fail because they cannot finish the trial.

6. Team

Drugs are built by people. (For now).

In human-developed pharma, the quality of a team shapes the quality of every decision that follows. Trial design. Regulatory strategy. CMC execution. Fundraising. Hiring. Partnering. Recovery after failure. None of this is random.

You start by mapping the people. CEO. CMO. CSO. Clinical lead. COO. Board. Advisors. Key investigators if they matter. But titles mean little by themselves. What matters is what these people have actually done.

That is where the work gets hard. The data is scattered. You have to connect individuals to prior companies, prior programs, prior outcomes. Approvals. Failures. Exits. Near misses. And you have to separate proximity from ownership. Being present during a successful program is not the same as driving it.

Then you look across a few hard questions. Has this team taken programs from preclinical to IND. From Phase 1 to Phase 2. From Phase 2 to Phase 3. From Phase 3 to approval. Each step has its own traps. Experience in one does not guarantee skill in the next.

Then relevance. Oncology experience does not automatically map to neurology. Small molecule experience does not automatically map to gene therapy. Early stage builders do not always become late stage operators. Commercial talent does not always help in preclinical strategy. Context matters.

Track record matters too. Are these repeat winners. Mixed operators. First timers. Failure is not disqualifying, but it has to be understood. What was their role. What did they learn. What pattern do they show.

Clinical and regulatory strength matter more than most people think. A strong CMO can rescue a program by designing the right trial. A weak one can kill it by getting the basics wrong. The same is true for regulatory leadership.

Then comes completeness. Are all the major functions covered. Clinical. Regulatory. CMC. Finance. Commercial if the stage requires it. Are there obvious holes. Is too much resting on one person. Are the advisors real contributors, or just names on a slide.

Scaling risk is real too. A team that is strong at the seed stage may struggle at Phase 3. A team that knows development may not know launch. Some teams grow with the asset. Some do not.

People are not interchangeable. Good systems need to know that. The probability of success depends, in part, on who is building the drug.

7. Regulatory

Regulatory is not magic.

It is history, made formal.

People often talk about approval as if it were opaque. It is not. Regulators are not improvising. They are applying precedent. Every approval teaches you something about the boundary of what is acceptable.

So you have to map that boundary. What drugs were approved in this indication. On what endpoints. In what populations. With what trial designs. Under what safety burden. Were the endpoints hard outcomes or surrogates. If they were surrogates, how established were they.

These are not background details. They are the rules of the game.

Then you compare. Does the current program align with that history, or drift away from it. If it drifts, is that a smart innovation or a dangerous gamble. An endpoint that has never supported approval may be visionary. It may also be dead on arrival. A trial design that departs from prior registrational standards may produce positive data that no agency wants to trust.

Alignment matters more than novelty here.

You also have to read the softer signals. Agency interactions matter, even when only partly disclosed. End of Phase 2 meetings matter. Designations matter too. Breakthrough Therapy. Fast Track. Orphan. RMAT. These are not just badges. They tell you how the agency sees the asset, the unmet need, and the path ahead.

Safety tolerance sits inside regulatory logic. Different diseases allow different levels of risk. Oncology is not obesity. Rare fatal disease is not chronic primary care. A system has to internalize those thresholds and measure the program against them.

And you have to think in failure modes. Could this end in a complete response letter. A demand for another trial. Rejection of the endpoint. Rejection of the effect size. Delay due to safety. Restriction to a narrow label.

Regulatory risk is rarely random. It is patterned. If you can model the pattern, you can often see the outcome before the meeting happens.

8. Manufacturing

A drug that cannot be made is not a drug.

That is the blunt truth of CMC.

This is where biology meets physics. It is where elegant science collides with yield, purity, consistency, storage, and scale. Many programs that look strong in papers break here.

Start with the modality. Small molecules. Biologics. Cell therapies. Gene therapies. RNA drugs. ADCs. They do not share the same manufacturing world. Some are robust. Some are fragile. Some are cheap to scale. Some are hard to make even once.

Then ask how mature the process is. Early processes move. Conditions shift. Methods evolve. Yields drift. Late stage processes should be locked, reproducible, and validated. The transition between those states is one of the most dangerous moments in development.

Scale is the next trap. A process that works in a lab may fail in a plant. Small batch success does not guarantee commercial success. At scale, throughput matters. Batch consistency matters. Tiny variations can change the product, especially in complex modalities.

Then comes change over time. If the process evolves between early and late trials, the product may not be the same. That creates comparability risk. You may end up with clinical data on one version and a commercial plan for another.

Quality systems decide whether the product can be trusted. Analytical methods. Release specs. Batch variability. Stability. Shelf life. Cold chain. Handling. A product that degrades too fast or varies too much is a problem no matter how strong the efficacy data is.

Supply chain matters more than people admit. Single source dependencies. Rare raw materials. Geographic concentration. Fragile vendors. All of these can break a program at the wrong time.

Cost matters too. A drug can work and still fail if it is too expensive to make. Cost of goods shapes margins, access, and pricing power.

CMC is not a box to check. It is a map of whether the product can exist in the real world. Reliable. Scalable. Economic. That is the test.

9. Intellectual property

Science alone does not create durable value.

If you cannot defend the asset, someone else will take the margin.

That is why IP matters so much. Not as legal decoration. As economic reality.

The first question is the strongest one. Do you own the molecule, or the construct. Composition of matter is the core. If that protection is broad and durable, it creates a real moat. If it is narrow or absent, competitors may be able to walk around it.

Then comes breadth. Does the protection extend across indications, formulations, dosing regimens, combinations, manufacturing methods. Strong portfolios do not rely on one claim. They layer protection. Weak ones leave doors open.

Time is as important as breadth. Patent life is finite. What matters is not the filing date alone, but how much runway remains at expected approval. A therapy that arrives with little protection left is in a very different position from one with a long exclusivity tail.

Then there is freedom to operate. A company can have patents and still be boxed in by someone else’s claims. Overlap matters. Blocking patents matter. Licenses matter. Settlements matter. You have to map the landscape, not just admire your own fence.

Litigation is part of the story too. Patents get challenged. Some collapse. Some survive. The likelihood of challenge, and the strength of the claims under pressure, matter a great deal.

And then there is practical defensibility. Even if the paper protection is there, how hard is it for a competitor to build something close enough to compete. Can they tweak the structure. Can they design around formulation. Can they enter with a biosimilar path.

IP is not static. It is a living system that decides how long an advantage can hold. Good biomedical intelligence has to read it with the same rigor it brings to clinical data.

10. Disease population and market size

This is where biology meets scale.

A drug may work. The next question is how many people it can really help, and what that is worth.

It starts with epidemiology. Incidence. Prevalence. Growth. But top line numbers mislead. The real work is in the cuts that come after.

How many patients are diagnosed. How many are treated. How many fit the label. How many meet the biomarker. How many reach the right line of therapy. Every filter shrinks the pool. What looks huge at first often becomes modest by the time you reach the true addressable population.

Segmentation matters. Diseases are not one thing. They have subtypes. Severity tiers. Biomarker subsets. Early line and late line use. A refractory niche is not a frontline market. A broad chronic disease is not the same as a narrowly defined genetic subset.

Geography matters too. Diagnosis rates differ. Treatment rates differ. Access differs. Reimbursement differs. A global number can hide a very uneven reality.

Growth matters. Some diseases expand because populations age, screening improves, or diagnosis gets better. Others do not. You cannot treat the market as static if it is moving under you.

Then comes the hard translation. Patient numbers into revenue. That means price. Penetration. Duration of therapy. Access. Those assumptions have to be explicit, because they drive the whole model. Small changes in uptake or pricing can swing the result far more than people expect.

This looks like math. It is really assumption management.

Most market models fail not because the arithmetic is wrong. Because the hidden assumptions are weak. Good systems make those assumptions visible and test how much they matter.

11. Commercial reality

Approval is not adoption.

Many approved drugs still fail. They fail because the world does not behave the way the slide deck said it would.

Commercial reality starts with differentiation. Is the drug clearly better than standard of care. Better efficacy. Better safety. Better convenience. Easier dosing. Easier administration. If the answer is only marginally yes, behavior may not change.

And behavior is the whole game.

Physicians have habits. Patients have fears. Payers have budgets. Hospitals have workflow. A product enters all of that at once.

Pricing is one part of it. A company may believe the clinical value supports a premium price. Payers may disagree. Then come step edits. Prior auth. Restricted access. Narrow coverage. Cost effectiveness thresholds. These are not downstream annoyances. They shape uptake from day one.

Practical friction matters just as much. Does the drug require biomarker testing. Does it need infusion centers. Is the dosing schedule awkward. Is monitoring intense. These things matter in the clinic far more than many models admit.

You also need launch benchmarks. What happened when similar drugs launched. How fast did they grow. Where did adoption stall. What access barriers appeared. Without that history, forecasts tend to become wishful thinking.

Competition changes the commercial picture too. A drug entering open space is one thing. A late entrant with modest differentiation is another. First is not always best. Best is not always enough. But timing always matters.

At the core, this dimension is about incentives and behavior. Physicians, patients, and payers each respond to different pressures. Any system that ignores that will overestimate uptake.

Commercial success is not just clinical value. It is clinical value filtered through the real world.

12. Competition

No drug exists alone.

Every program lives inside a moving field of other programs. Approved drugs. Late stage challengers. Early stage threats. Same mechanism. Different mechanism. Better safety. Better timing. Better convenience. All of it matters.

So the first job is mapping the field. What is standard of care. What is approved. What is in Phase 1, 2, and 3. What mechanisms are in play. What modalities. What patient groups. What readouts are coming.

But mapping is not enough. You have to compare.

Direct competitors matter most when they hit the same pathway. That is where class effects emerge. One program’s failure can poison the well. One program’s success can validate the whole area.

Indirect competitors matter too. A new modality can make the old frame obsolete. Gene therapy can reset the bar for chronic drugs. A safer oral can damage an infusion franchise. A different mechanism can win if it solves the problem better.

Timing is brutal here. Who reads out first. Who reaches the market first. Who comes later with better data. A program that looks strong today can be leapfrogged before launch.

So differentiation has to be judged in context. Better efficacy. Better safety. Better convenience. Something that matters. If the difference is thin, the drug risks becoming a me too product in a crowded field. If the difference is real, it may become best in class. But that judgment cannot be static. The field keeps moving.

Market structure matters too. Some spaces can support many winners. Others collapse into a few dominant players. You need to know which kind of market you are in.

History helps. Every therapeutic area has a pattern. Some mechanisms keep working. Others keep disappointing. Some strategies win early and hold. Others get overtaken. A good system learns those patterns and uses them.

Competition changes the meaning of everything else. Strong data may not be enough if someone else is stronger. Middling data may still win if the field is weak. Nothing is judged in isolation.

That is why competition is the final dimension. It turns every absolute into a relative one.

Why these twelve matter together

What makes these dimensions powerful is not just what each one says. It is how they interact.

A strong mechanism can be ruined by weak protocol design. Good efficacy can be capped by safety. A well run program can still lose on enrollment. A good drug can fail in the market. A strong asset can be boxed in by weak IP. A promising program can be overtaken by competition.

This is the real challenge for clinical trial intelligence.

It is not enough to retrieve papers and summarize them. The system has to do four harder things. It has to retrieve the right primary data. It has to structure and compare that data. It has to judge what matters. And it has to develop taste.

That last part matters most.

Retrieval is getting solved. Analysis is getting solved.

Judgment is not.

Taste barely exists.

And taste is what tells you when a result feels brittle. When a protocol is optimized to win. When a management team is out of its depth. When a safety signal will widen. When a market model is built on fantasy. When a program that looks strong on paper is weaker than it seems.

That is why current systems still fail in high stakes biomedical work. They can find. They can summarize. They can rank. They still struggle to know.

The path forward is not just bigger models. It is better framing.

These twelve dimensions offer one such frame. They define the major decisions that shape drug development. They force the right questions. They create structure where people often rely on instinct alone.

Miss one dimension and the program breaks.

Not always at once. Not always loudly. But predictably.

Capture all twelve, and you begin to see the system the way the best drug developers do.

That is the bar.

If we want to understand how America maintains or loses its edge to China and others, we need a clear framework. I think about US biotech leadership across four levers:

1. Science

2. Commercial

3. Regulatory

4. Operational

Each lever has subcomponents. Each can either compound the U.S. advantage or quietly erode it.

Below is how I see the system and where policy and technology can make the biggest difference.

Preclinical Science: The Foundation

Panoptic Bio is a pharma intelligence company in Mountain View, CA

The American biotech advantage starts in universities.

The U.S. system of government funded preclinical research, primarily through NIH, NSF, SBIR, and related programs, has no true peer. Add to that DARPA, ARPA H, and defense linked funding streams, and you have a deep reservoir of early stage discovery science.

This is not an accident. It is a policy choice.

Preclinical science is hard to fully privatize. The timelines are long. The failure rates are high. The ROI is uncertain. Venture capital does not fund decades long fundamental biology programs without a translational hook.

Later stage development can be privatized. Early stage science cannot be fully privatized without hollowing out the pipeline.

If the United States cuts government science funding, it is not trimming fat. It is weakening the root system of its pharmaceutical tree.

China understands this. It is massively expanding state backed fundamental research. If the U.S. cuts while China expands, the long term consequence is obvious.

The U.S. and China are in a strategic competition with areas of collaboration to advance biotechnology and pharma innovation, deciding who will lead in the 21st Century.

It is critical the U.S. and Europe preserve and expand federal funding for preclinical science.

If politically difficult, there is a second path.

The U.S. should treat biotech as national security.

Public health is national security. Healthy civilians and healthy soldiers are strategic assets. Biosecurity is a dual use domain. Just as climate was once framed as a defense issue, biotechnology should be embedded more deeply into defense budgets.

If civilian science funding contracts, defense aligned funding should expand to compensate. The Military can play a larger role in sustaining American bio innovation capacity through its funding vehicles.

Immigration policy also matters here. The US academic system attracts global talent. If funding shrinks and visas tighten, the U.S. will hollow out its own preclinical advantage.

Preclinical science is not optional. It is the seed corn.

Clinical Stage Science & Capital Markets

Once assets move into the clinic, the funding structure changes.

Here the United States has a powerful advantage: capital markets.

VC. Growth equity. Private equity. IPO markets. Follow on offerings. Private placements. Structured credit. Pharma partnerships.

The US has deep, risk tolerant capital pools that fund clinical development at scale. China has capital, but it does not have the same depth of global commercial opportunity or exit liquidity.

But capital allocation matters.

The U.S. should not only fund therapeutic assets. The U.S. should fund infrastructure:

• AI enabled drug discovery platforms

• Genetic and multiomics infrastructure

• Trial technology

• Data platforms

• Manufacturing innovation

Private capital should increasingly flow into the infrastructure that accelerates all future therapeutics, not only the next molecule.

Commercial: Pricing and Global Fairness

The commercial lever is about pricing and incentives.

The United States currently supports biotech innovation through high domestic pricing. US patients often pay two to ten times what patients in Europe pay for the same drug.

In practice, America subsidizes global innovation.

Other countries regulate pricing tightly. They wait for US funded innovation and then negotiate aggressively. This is a form of pharmaceutical protectionism.

If the US maintains high pricing domestically but does not push for market openness abroad, American patients shoulder disproportionate burden.

The answer is not simply to crush US pricing. That would damage innovation incentives.

The better path has two components:

1. Preserve domestic pricing incentives that sustain ROI.

2. Pressure international markets to open access.
   

Most favored nation pricing proposals are politically attractive. But if applied without higher foreign pricing, they effectively import low price regimes into the US and compress global innovation margins.

If we want fairness, we must address global pricing asymmetry, not only domestic optics.

Rebates and Trial Location Arbitrage

Commercial policy also affects where trials happen.

US clinical trials are extraordinarily expensive. CRO dominated infrastructure, administrative burden, and site costs push per patient costs into six figures in some indications.

In contrast:

• China can run trials at a fraction of the cost.

• Australia offers substantial government rebates, sometimes around 40 percent.

• Other jurisdictions aggressively compete for clinical trial flow.

If the US wants to reshore trials, it needs incentives. Rebates tied to domestic clinical trial spending could bring trials back onshore.

Otherwise, rational companies will continue to run pivotal programs abroad and bridge into the US later.

Regulatory: Safety, Speed, and Statistical Innovation

The FDA, EMA, and NMPA represent three major regulatory centers of gravity.

Historically, the FDA has been the most rigorous and often the most difficult to navigate. In many therapeutic areas, sponsors perceive it as more conservative than the EMA. The NMPA has been less rigorous historically but is increasing standards.

Regulatory stringency is not inherently bad. Safety matters. But overregulation can silently cost lives through delayed approvals.

We need an honest conversation:

• How many lives has the FDA saved by not approving unsafe drugs?

• How many lives has been lost because beneficial drugs were delayed or never approved?
  

We do not measure the second number clearly. That is a policy blind spot.

Reforms worth expanding include:

• Acceptance of a single pivotal trial in certain oncology settings.

• Greater use of real world evidence.

• Broader adoption of Bayesian statistical methods in SAP design.

• Adaptive and accelerated pathways that balance uncertainty with urgency.

The regulatory goal should not be deregulation for its own sake. It should be optimized risk balancing.

We must balance patient safety today with patient health tomorrow.

If Europe is easier to get through than the United States, and if China continues accelerating its pathways, the US risks becoming the slowest node in the global innovation chain.

That is not compatible with leadership.

Operational: Trials, Data, and Infrastructure

This is where much of the hidden inefficiency lives.

Clinical Trial Cost Structure

CROs dominate the current system. Trials are centralized. Sites are dependent. Administrative overhead is immense.

Enrollment delays kill companies. Companies run out of runway because recruitment lags.

There is another model.

Decentralize trials.

• Physician led trials

• Professor led academic trials

• Enabling community clinics as sites

• Reducing dependence on centralized CRO control
  

In China, physician autonomy in trial participation is broader. If every qualified physician office in the US could more easily function as a trial site, enrollment velocity would increase dramatically.

We need infrastructure that lowers friction to becoming a site.

Heterogeneous Populations as Strategic Advantage

The US population is heterogeneous. That is an advantage.

Chinese data often requires bridging studies before being accepted in Western markets because of population homogeneity concerns.

US trials, if designed with broad inclusion criteria and heterogeneous enrollment, can generate globally generalizable data.

We should codify heterogeneous representation requirements not as ideology but as global competitiveness strategy.

If American trials reflect the world, American drugs become global by default.

Data Infrastructure: Fragmented vs Unified

China benefits from unified medical data infrastructure. It can query large portions of national health data, identify rare disease patients, and accelerate enrollment.

The United States has:

• Fragmented EHR systems

• State level legal variation

• Privacy complexity

• Limited interoperability

• Barriers to large scale model training
  

We cannot ignore this.

Unified does not mean unsafe. Encrypted, privacy preserving, federated architectures can enable large scale AI without exposing raw data.

But if we do not build national scale health data infrastructure:

• Patient to site matching remains inefficient.

• Rare disease identification remains slow.

• Trial recruitment remains costly.

• AI models remain underpowered.

Data is the substrate of modern biotech.

Clinical Trial Intelligence and Workflow Automation

The next frontier is workflow automation across the entire biotech lifecycle.

Signal capture and due diligence
Patient matching and enrollment
Dosing and follow up
Manufacturing and CMC
Quality control
Regulatory documentation
Commercial launch
Marketing

Every one of these is a workflow problem.

Clinical trial intelligence means:

• Automating patient identification across EHR systems.

• Automating eligibility checks against inclusion and exclusion criteria.

• Automating site activation.

• Automating data monitoring.

• Automating safety signal detection.

• Automating CMC documentation.

• Automating market access analytics.
  

This is not only cost reduction. It is speed. And speed saves lives.

If AI is trained on large scale, well structured, interoperable health data, the US can create a compounding advantage in development velocity.

If we fail to modernize data infrastructure, others will.

The Strategic Picture

Across science, commercial, regulatory, and operational levers, the pattern is clear:

• Expand and protect preclinical government funding.

• Leverage defense budgets for biosecurity and dual use biotech.

• Preserve domestic pricing incentives while pushing for international fairness.

• Modernize regulatory pathways with adaptive and Bayesian methods.

• Reduce trial costs through decentralization and site empowerment.

• Build unified, privacy preserving health data infrastructure

• Invest in clinical trial intelligence and workflow automation.

Biotech leadership is not a single policy decision. It is a systems outcome.

China is a gift to the United States. In the sense that competition is exactly the urgency the pharmaceutical industry needs to wake up and become more efficient.

Clinical Trial Intelligence plays a key role in this drive towards greater efficiency.

Clinical trial intelligence is the integrated, AI driven layer that sits across the entire clinical development process, structuring fragmented data into a unified model of trial risk and decision making. It pulls together scientific rationale, protocol design, regulatory precedent, operational performance, competitive landscape, CMC readiness, and commercial viability into a single probabilistic framework. Instead of leaving these domains siloed across PDFs and functional teams, clinical trial intelligence transforms them into a structured system that benchmarks against historical programs, runs counterfactual scenarios, and identifies where failure is most likely to emerge before it happens.

Why Panoptic Bio Exists

This is the gap Panoptic Bio addresses.

If efficiency is the constraint, and Phase I to approval remains a roughly 7.9 percent outcome over more than a decade, the biggest leverage is not marginally better discovery. It is a better clinical-stage decision system.

Faster decision cycles. Higher-quality evidence synthesis. Greater consistency. Scalable learning across programs.

Clinical trials should not be driven by fragmented, manual intelligence. They should be guided by systems that learn continuously from trial reality and translate that learning into better decisions and better execution.

This is where failure is most expensive. It is also where decision intelligence matters most.

If China’s advantage is execution throughput, the Western response must be decision superiority paired with execution strength. That means building a modern pharma operating system that converts AI leadership into sustained biotech leadership.

Panoptic Bio’s Trial Terminal is a decision intelligence platform to safely run AI models and agents on ongoing clinical trials

Retrieve any registered clinical trial from the U.S., Europe, China, or globally

Run Probability of Success analysis on ongoing clinical trials

Who will lead the 21st century?

The importance of clinical trial intelligence lies in addressing the real bottleneck in pharma, which is not molecule discovery but the 92 percent failure rate in clinical development. Even strong biology fails when one weak dimension cascades across the system. By quantifying risk, forecasting outcomes, and recommending targeted interventions at the design stage before millions are spent, clinical trial intelligence increases the probability that programs reach approval. Even a one percent improvement in success rates translates into enormous economic value and, more importantly, more therapies reaching patients. In this model, clinical trial intelligence becomes the infrastructure layer of an AI native pharma stack, determining which discoveries ultimately become medicines.

In the end, biotech leadership will not be decided solely by who discovers the most novel targets, but by who builds the fastest and most intelligent development system. The United States still leads in science and AI, but without upgrading its decision infrastructure, that advantage will dissipate into inefficiency. China’s rise demonstrates that execution throughput compounds into strategic leverage. The next era of pharma will be defined by integrated decision intelligence that shortens learning cycles, improves capital allocation, and increases the probability that the right programs reach patients. Sovereignty, competitiveness, and patient impact now hinge on who builds the better decision loop. Bio leadership in the 21st century belongs to those who deploy clinical trial intelligence.

The West will not “science its way” to biotech leadership unless it rebuilds the drug development engine with the same intensity applied to discovery. China’s rise shows what happens when execution capacity compounds.

The United States still leads in frontier AI and foundational science. But without major gains efficiency, that advantage will not translate into durable health leadership or medical sovereignty.

At the center of the efficiency gap is decision intelligence.

The Problem Is Not Just Biology

The headline numbers are familiar, which is why they are dangerous.

Drug development now averages roughly $2.6 billion per approved drug in capitalized cost, based on Tufts CSDD estimates. Time from first-in-human to approval averages about 10.5 years, according to BIO and Informa.

These figures imply something uncomfortable. Even with perfect biology, the system would remain slow and expensive. Too much failure and capital loss is created by how decisions are made.

The industry has pushed hard on discovery. AI for Science and AI for Bio are now standard. Generative design, target discovery, structure prediction, and self-driving labs have advanced quickly. But better candidates do not guarantee better outcomes.

Discovery gains are routinely lost when diligence is manual, portfolio decisions lag data, trials are slow to design, competitive intelligence is fragmented, endpoints are poorly specified, sites under-enroll, protocol amendments multiply, and regulatory strategy becomes reactive.

BIO and Informa data show where this breaks down. The probability of a Phase I asset reaching approval is roughly 7.9 percent across 2011 to 2020.

Phase II is the largest bottleneck. Only about 28.9 percent of candidates advance. That is not just biology. It reflects late, noisy learning.

Discovery AI increases the number of bets. It does not ensure better bets. The West still runs programs through slow, fragmented decision systems. The result is not only execution risk. It is low decision quality and the resulting probability of success variability.

China’s Advantage Is Fast Learning

Calling China biotech’s “back office” understates the shift. China did not just build molecules. It built the industrial throughput layer of drug development. Over time, that layer became strategic leverage.

The licensing data is clear. A Boston Consulting Group (BCG) reports that deals involving China drove almost half of global biopharma licensing activity in 2025. Valuations of Chinese assets rose roughly 150 percent year over year, outpacing both US and European peers.

Boston Consulting Group (BCG) analysis on pharma deal value reveals China’s rise

China’s share of global biopharma licensing value grew from about 5 percent in 2021 to roughly 40 to 50 percent by 2025.

In January 2026, AstraZeneca signed a multi-program obesity deal with CSPC valued at up to $18.5 billion, including a $1.2 billion upfront payment. It was one of the largest licensing deals of the 2025 to 2026 period.

China is no longer just lower-cost execution. It is a source of assets and a high-throughput development environment that global pharma depends on.

This advantage came from sustained investment in the execution stack. Large-scale CRO and CDMO capacity turned scientific intent into executable trials. Firms adapted under geopolitical pressure by restructuring and focusing on core platforms. That is industrial maturity.

The real compounding advantage is cycle time. Cycle time determines when decisions are made.

Faster iteration means earlier learning, faster termination of weak programs, tighter capital concentration, and stronger partner pull. This is how a back office becomes a front office.

At its core, China built a fast learning system.

Sovereignty Depends on Decision Loops

The West has already seen efficiency collide with sovereignty in consumer technology. TikTok triggered years of political conflict in the US. China blocked US social platforms much earlier.

Pharma has had its own preview. COVID made medical sovereignty tangible. Countries with domestic development and manufacturing capacity moved first. Others waited.

If algorithms and chips are treated as sovereignty concerns, medicines will be too. Possibly more so.

Medical sovereignty is not only about factories. It is about decision autonomy. The ability to design trials, interpret data, adjust strategy, and advance or kill programs without dependence.

Decision loops are sovereignty.

Collaboration Will Persist

Decoupling is not clean. US-China biopharma collaboration remains deep and, in some areas, growing.

Pharma needs assets. Biotech needs capital and commercialization. Patients need therapies. These pressures are structural.

Pharma faces a patent cliff. Populations are aging. Healthcare demand is rising. Biotech has faced a funding winter since 2021.

Deal data reflects this reality. China’s rising licensing share is proof.

Two things are true. Innovation is global, and collaboration will continue. At the same time, the US and Europe want operational self-sufficiency in critical medicines.

This tension defines the decade. And it leads back to efficiency.

Efficiency Starts with Decision Intelligence

Start with the baseline. $2.6 billion. 10.5 years. 7.9 percent approval probability.

Under these constraints, advantage will not come from discovery alone. It will come from better decision intelligence paired with strong execution.

Discovery creates options. Decision intelligence determines which options deserve capital, time, and operational focus.

The real levers now are earlier selection of the winning 8 percent, real-time integration of biological, clinical, regulatory, and commercial signals, shorter decision cycles, faster trial design, higher signal-to-noise in endpoints, and tighter control of sites and amendments.

AI must extend beyond discovery. Models can propose molecules. The larger prize is compressing and de-risking the clinical and regulatory arc, where most capital is lost.

Most capital is not lost because molecules fail. It is lost because failure is discovered too late.

Turning the US AI Edge Into Pharma Outcomes

The United States retains major advantages. Frontier AI research. Large-scale compute. Deep capital markets. Leading academic medical centers. Influential regulators.

But advantages only matter if they are operationalized. They transfer through tools and workflows that improve everyday decisions.

AlphaFold mattered because it collapsed iteration time and reshaped workflows. The same must happen in clinical development.

AI must become infrastructure. Systems that continuously synthesize evidence and guide decisions in real time.

Given China’s efficiency gains, the US does not have time to wait.

Why Panoptic Bio Exists

This is the gap Panoptic Bio addresses.

If efficiency is the constraint, and Phase I to approval remains a roughly 7.9 percent outcome over more than a decade, the biggest leverage is not marginally better discovery. It is a better clinical-stage decision system.

Faster decision cycles. Higher-quality evidence synthesis. Greater consistency. Scalable learning across programs.

Clinical trials should not be driven by fragmented, manual intelligence. They should be guided by systems that learn continuously from trial reality and translate that learning into better decisions and better execution.

This is where failure is most expensive. It is also where decision intelligence matters most.

If China’s advantage is execution throughput, the Western response must be decision superiority paired with execution strength. That means building a modern pharma operating system that converts AI leadership into sustained biotech leadership.

Panoptic Bio’s Trial Terminal is a decision intelligence platform to safely run AI models and agents on ongoing clinical trials

Retrieve any registered clinical trial from the U.S., Europe, China, or globally

Run Probability of Success analysis on ongoing clinical trials

The Leadership Question

Biotech has long told itself that biology is the bottleneck.

Biology is hard. But the numbers show another constraint. How decisions are made.

China’s rise from back office to strategic center is both warning and lesson. Execution compounds. But execution follows decisions.

If the West wants to lead scientifically, commercially, and geopolitically, it must compete where outcomes are determined.

Decision intelligence is now the frontier.

Sources
Panoptic Bio: https://www.panoptic.bio
Forbes: https://www.forbes.com/sites/shimiteobialo/2026/01/21/how-china-became-the-biotech-industrys-back-office/
BCG: https://www.bcg.com/publications/2026/reimagining-business-models-biopharma-trends
GlobalData: https://www.globaldata.com/media/business-fundamentals/large-pharma-drug-licensing-from-china-reaches-record-high-at-28-in-2024-reveals-globaldata/

On the widely used TOP/HINT benchmark, evaluated using the same train/test split and evaluation protocol as published baselines, our Panoptic Bio model achieved an ROC–AUC of 0.76.

Our result reflects improved discrimination on late-stage trials under a fixed evaluation metric.

OpenAI and Babylon Biosciences

This year, media reports described a collaboration between OpenAI and Babylon Biosciences in which a clinical trial prediction model trained on 430 historical clinical trials reported an AUC of approximately 0.84. These results have not appeared in a peer-reviewed publication and were reported through media coverage. It is also important to note that the dataset of 430 clinical trials was not published, and that there are significant differences in task definition and evaluation metrics compared to the TOP/HINT benchmark.

Panoptic Bio’s evaluation was conducted on a substantially larger corpus, spanning:

• 11,600 historical clinical trials

• 560,398 unique documents, including scientific, regulatory, commercial, and operational sources

Predictions are produced as calibrated probabilities with accompanying uncertainty estimates (including confidence intervals and precision metrics), making outputs suitable for direct use in due diligence, portfolio analysis, and go/no-go decisions in clinical development, an area where more than $276B is spent annually.

Access is available through the Panoptic Bio Trial Terminal.

What We Are Sharing vs Not Sharing

We share here:

• The definition of the prediction target

• The evaluation setup

• Benchmark results

We are not publishing:

• The full feature set

• Training recipes

• Our proprietary system architecture details

We are happy to share further technical evidence with qualified partners and to evaluate externally specified trials so results can be independently verified.

Try It on Your Trial

In the spirit of the holidays, we are offering a limited number of trial evaluations at no cost for select partners in the Panoptic Bio Trial Terminal.

Consider it a free trial. No pun intended.

A Peak Inside the Panoptic Bio Trial Terminal

Our team has worked tirelessly on this. Our lead applied AI scientist was trained by two of the most prominent researchers in the field who created a novel, now famous, lightweight ensemble model. Much gratitude to this milestone and those in our team who supported this early result.

Panoptic Bio, Inc is a pharma intelligence company based in Mountain View, CA

The first pharma agent will not discover a molecule. It will not replace a scientist. It will not run a company. It will do something much smaller and much more important. It will read.

The real bottleneck

Pharma today is not limited by intelligence. It is limited by fragmentation.

A single clinical program generates tens of thousands of documents. Protocols. SAPs. Clinical study reports. Regulatory letters. Internal decks. They live in different systems, owned by different teams. No single human ever sees the full picture.

The cost is real. Clinical teams spend 30 to 40 percent of their time searching for and reconciling information. Not analyzing it. Senior leaders rely on summaries that are outdated, filtered, and incomplete. Critical assumptions are lost or contradicted without being noticed.

A Day in the Life of a Biotech Leader isn’t Pretty:

A reader and a unifier

The first pharma agent is a reader and unifier. More precisely, an analyst for due diligence. Its role is not creativity or discovery, but judgment support at the moments that matter most in clinical development.

It does not replace decision makers. It improves the quality of their decisions.

The agent begins with relevance driven retrieval, identifying the right trials, comparators, and historical precedents. It then reads across the full program as a system spanning science, trial design, regulation, operations, and competition, focusing not on volume but on alignment.

Dimensional Analysis

This analysis is structured through the twelve dimensions of clinical trial success. Mechanism of action, efficacy, safety, and disease population form the scientific core. Protocol quality, enrollment strength, and clinical team capability define the operational center. Regulatory path, CMC readiness, intellectual property, competitive landscape, and commercial viability connect the program to the external world.

Each dimension matters on its own. What matters more is how they interact.

A workflow of agents concurrently calls the core database retrieves dimension specific data and runs a cascade of analytical algorithms to score it.

From structure to score

The agent does not turn this into a narrative. It turns it into scores. Each dimension is evaluated independently, then benchmarked against an underlying dataset of real programs. Successes. Failures. Near misses. False confidence. Unexpected wins.

Weaknesses surface even when the commercial opportunity looks strong.

From this structured analysis comes the output every investment committee, portfolio review, and development decision ultimately seeks. Probability of success.

This is the core due diligence task in pharma. Everything else follows from it.

Once probability is established, value can be modeled. The agent performs market analysis, builds pricing and adoption assumptions, constructs discounted cash flow models, and runs sensitivity analyses.

It produces the same artifacts humans already use. Models. Spreadsheets. Explicit assumptions. Transparent logic. Nothing is hidden. Nothing is implied.

Human judgment remains

Humans remain fully accountable. They challenge the inputs. They adjust the assumptions. They override the conclusions when judgment demands it.

What changes is not who decides, but how informed that decision is.

This is the first pharma agent. Not a scientist. Not an executive. A due diligence agent. This is where autonomous pharma actually begins.

How it all comes together:

In 10-15 years, pharma will look like this.

As a Full Stack AI Pharma Company, the majority of core functions of clinical trials will be run by specialized AI agents across science, regulatory, clinical, and commercial workflows. Each agent handles a distinct domain, from mechanism-of-action analysis to protocol design to CMC and portfolio strategy. Together they operate as an autonomous layer that reduces risk, and scales pharma value for patients.

Only 7 human leaders will be needed to run a $100B pharma.

Their workflows will not be fully run by agents, but rather with agents. The key is that these workflows must be automatable. Each step in a clinical trial follows a pattern that can be learned, modeled, and executed by an agent that specializes in that domain. Human leaders set direction, make final judgments, and guide strategy, while agents perform the continuous analysis, monitoring, forecasting, and decision support that no human team can sustain at scale. The agents do the work that is repeatable and data driven. The humans do the work that is creative and strategic. Together they form a new operating model for pharma.

Currently, the majority of the focus in AI for Pharma is not here. It is in drug discovery. Everyone I know is in this space. It is like a gold rush for gold.

You can see how little white space exists in discovery and preclinical. When I say white space, I quite literally mean white space on the page of the market map!

Yes, the AI drug discovery boom has created a new problem. Discovery is now crowded. Many great molecules are here already, and many great ones are coming as dozens of platforms compete intensely across every therapeutic area. Oncology, neurology, autoimmune, infectious disease, metabolic, rare disease, and more will see a new revolution.

Yet even with all this innovation, good molecules still fail in clinical trials. This is why a new layer. One that determines which molecules actually make it to patients. And helps the molecules go through the clinical stage, not just discovery stage.

It is true that only 8% of clinical trial programs take a drug to market. It takes on average 10 years and $2.5B end to end. When it works, it works big:

But these are the rare 8% that succeed. Every domain and dimension has to line up. The domains of science, regulatory, commercial, and trial operations have to be

flawless. And the twelve dimensions have to work out.

The twelve dimensions of a clinical trial capture the full reality of drug development. They span the scientific core, such as mechanism of action, efficacy, safety, disease population, and CMC. They include the operational center of gravity, such as protocol quality, enrollment strength, and the capability of the clinical team. And they extend into the external world, such as regulatory strategy, commercial viability, intellectual property strength, and the competitive landscape. These dimensions form the true map of risk. Each one can move a program forward or hold it back. Each one matters.

What makes these twelve dimensions powerful is their interdependence. A program can have strong biology but weak enrollment. It can have a strong protocol but an unclear regulatory path. It can have encouraging efficacy signals but face a crowded competitive field. A single weakness can cascade across the rest of the system until it limits the probability of success. This is why trial outcomes are rarely the result of a single factor. They emerge from the interaction of all twelve. Only when every dimension aligns do programs rise into the small group that reaches patients.

A weakness in any one of these areas can cascade into failure, even when the science and team are strong. This is why one mistake in one dimension is often enough to put a program into the 92% of trials that fail.

Pharma is actually changing much faster than people think, and much further beyond early stage drug discovery. The clinical stage is changing, too.

And the reason is simple: the problem in drug development is no longer the lack of data; it’s the lack of integrated intelligence to turn that data into decisions.

Clinical trial intelligence 101
Clinical trial intelligence creates a complete and structured view of a trial. It pulls together every document, every past study, every regulatory precedent, every biomarker dataset, every safety signal, every competitive readout, every dosing pattern, and every patient population variable. It then transforms this chaos into a unified and interpretable model of the trial. Instead of scattered PDFs and siloed experts, teams get an integrated map of risk across science, clinical strategy, regulatory pathways, and commercial positioning. Clinical trial intelligence is what finally connects the dots.

Once the data is structured, the system begins to reason. It analyzes each of the twelve dimensions of trial success, quantifies their strength, predicts where failure is most likely to emerge, and benchmarks every decision against historical analogs across thousands of programs. It runs counterfactual scenarios. How would success probabilities change if inclusion criteria shifted. If endpoints changed. If CMC timelines slipped. If the competitive landscape evolved mid trial. This gives leaders something they have never had before. Early visibility into failure before it happens, and a probabilistic explanation of why.

And finally, clinical trial intelligence intervenes. It recommends specific protocol refinements, enrollment adjustments, safety mitigations, statistical design improvements, regulatory strategies, site mixes, and even portfolio sequencing. It identifies the weakest dimension, explains the root cause, and suggests the corrective action with the highest expected value. Instead of discovering problems at interim analysis after millions are spent, trial intelligence surfaces them at the design stage when they are still solvable. The outcome is simple but transformative. Fewer preventable failures, more successful programs, and more molecules that actually make it to patients.

A new stack is emerging.

And the impact will be enourmous.

The power of one percent
A one percent increase in clinical trial success may sound small, but in the $1.7 Trillion pharmaceuticals industry it is indeed enormous. One percent more programs reaching approval translates into one hundred seventy billion dollars in new annual value for the industry. It also means more therapies reaching the people who need them. For cancer patients, for people with neurodegenerative disease, for families living with rare disorders, that one percent is not a statistic. It is time. It is survival. It is the difference between a pipeline that stalls and a pipeline that delivers. A one percent improvement reshapes the economics of pharma, but it also reshapes lives.

The real winners in the gold rush
The current boom in AI for drug discovery feels a lot like a gold rush. Everyone is rushing into the mountains, digging for new molecules, believing that the next strike will change everything. But in every gold rush, the real winners were the ones who looked ahead to the next part of the value chain. The ones who built the railroads, the banks, the equipment, the infrastructure that turned raw discoveries into lasting value. Discovery is the digging. Clinical trial intelligence is the infrastructure that determines which discoveries become real medicines.

In 10-15 years, pharma will look like this. The advent of autonomous pharma can be accelerated by about 5 years at most by the right company, the right team, and the right set of decisions. It will inevitably by here by 2040. But it can come sooner in the 2030s with the right push by the right people.

****

Wong, C. H., Siah, K. W., & Lo, A. W. (2019). Estimation of clinical trial success rates and related parameters. Biostatistics.
https://doi.org/10.1093/biostatistics/kxx069

DiMasi, J. A., Grabowski, H. G., & Hansen, R. W. (2016). Innovation in the pharmaceutical industry: New estimates of R&D costs. Journal of Health Economics.
https://doi.org/10.1016/j.jhealeco.2016.01.012

We do not know its exact form yet, but we know its general outline. The pattern is familiar. The pressure is rising. Viruses cross into humans all the time; most fail, some linger, a few adapt. It takes only one with the right properties to set the world on fire. It could be another coronavirus, a Covid 3, produced by the same evolutionary engine that gave us SARS in 2003 and COVID 19 in 2020. It could be a new respiratory RNA virus that moves efficiently through air and aerosol, the kind that spreads before symptoms appear and punishes any delay in detection.

It will almost certainly be viral, not bacterial. Bacteria announce themselves. They are slower, heavier, easier to treat. Viruses slip through surveillance. They mutate fast. They exploit our mobility, our density, our networks. It will almost certainly be zoonotic. That is how pandemics begin: the boundary between humans and animals thins, a virus jumps, adapts, learns the human body, and then learns how to move from one human body to the next.

It could be a novel influenza A strain from the H5, H7, or H9 lineages, the strains we monitor constantly because they look like threats waiting for an opportunity. It could be a paramyxovirus like Nipah or Hendra, both capable of high mortality and silent spread before symptoms. It could be an arbovirus like dengue or Zika, which already move through massive populations and could evolve into more efficient respiratory or vertical transmission.

These possibilities are not remote. They are not speculative. They are not doomsday fiction. They are the baseline. The world now lives in an era where spillover events happen more often, climate pushes species together, and global travel turns a single infection into a global footprint within days. The next pandemic is not a surprise. It is a schedule we have not seen yet.

The Missing Quadrant

We were fortunate during COVID 19. It was highly contagious but less deadly. Ebola is the mirror image: highly deadly but not very contagious. The catastrophic quadrant is the one we have not experienced.

The Big One

A virus with the transmissibility of Covid and the mortality rate of Ebola, or HIV, would be an event of global destabilization, a billion deaths in a single arc. The bubonic plague scaled to a connected world. An airborne HIV.

Pandemics usually begin with a simple and brutal truth. Viruses mutate at random. Most mutations do nothing. Some weaken the virus. A few give it an advantage. When those rare mutations appear inside an animal host and allow the virus to jump species, we get zoonotic spillover. Bats, birds, pigs, primates, and countless other reservoirs carry vast viral libraries that evolve constantly. When a mutation gives a virus the ability to enter human cells, replicate efficiently, or spread through respiratory droplets, the spark is lit. The jump is accidental. The consequences are not. This is how pandemics start.

Another new and deeply concerning pandemic risk is the possibility of human designed viruses created with the help of advanced AI systems. The danger is not that AI magically invents pathogens on its own, but that it can accelerate scientific workflows, lower barriers to complex biological design, and help inexperienced or malicious actors navigate what used to be extremely difficult steps. This creates a world where tailored viruses could, in theory, be engineered with properties that natural evolution rarely combines at once: high transmissibility, delayed symptoms, immune evasion. As AI becomes more capable, society must treat engineered pathogens as a new class of threat, one that arises not from random mutation or animal spillover but from human intent amplified by powerful tools. Even discussing this requires great care. One should not in detail describe methods, but rather name the risk clearly. But this must be discussed. Another essay in the near future will be published here. This essay however, focuses on pandemic responses and what clinical trial intelligence could mean in preparing for the next pandemic.

The World in 1918 vs 2020

The Spanish Flu of 1918 lasted nearly two years. It infected half a billion people and killed tens of millions. No vaccine existed then, not during the pandemic and not for decades afterward. The scientific world of 1918 had never seen a virus under a microscope. The electron microscope would not exist until 1931. Many scientists targeted “Pfeiffer’s bacillus,” believing incorrectly that a bacterium caused the disease. There were no genetic tools, no cell culture, no antivirals, no antibiotics, and no real virology. World War I dismantled laboratories and public health infrastructure. Humanity faced a virus of extraordinary lethality with almost no understanding of what it was fighting.

A century later, COVID 19 unfolded in a world with a completely different scientific arsenal. Chinese scientists sequenced the SARS COV 2 genome within 48 hours and posted it publicly on January 10, 2020. mRNA vaccine design began that same weekend. Two decades of platform work on lipids, mRNA stability, and earlier coronavirus vaccines suddenly activated. The global scientific machine spun up: billions of dollars, thousands of trial sites, rolling reviews at the FDA, real time data infrastructure across continents. It produced a vaccine in eleven months, the fastest in human history. It was astonishing. And still, the process was messy.

Complexity

Trials chased a moving virus: Wuhan strain, then Alpha, then Delta, then Omicron. Every shift changed the underlying biology of the target. A vaccine trial designed for one variant would face a different one by the time endpoints arrived. Human immune systems layered more variability on top: age, conditions, genetics, prior infections, behavior. Trials required real infections to measure efficacy, which meant sites had to follow the geography of outbreaks in real time. Ethics demanded that placebo participants be vaccinated once early data showed benefit, which constrained long term insights. Rare side effects required global surveillance with sample sizes in the millions. Manufacturing had to begin while trials were still underway, burning billions of dollars before approval. Public distrust, misinformation, and political pressure warped enrollment and exposure patterns in ways no model could predict. Biology is complex. Humans are complex. Pandemics move faster than clinical research.

This is the environment biotech leaders operate in. A biotech CEO lives at the intersection of uncertainty, incomplete information, and irreversible decisions. Few professions require integrating science, business, operations, and regulation under this much pressure. Every choice is a bet made with partial data. I learned this firsthand leading Healis Therapeutics, becoming the youngest cofounder and leader of a clinical stage neuroscience company. I had to make decisions that even seasoned executives lose sleep over. Which indication should we prioritize. How will regulators interpret our design. What signal matters most to patients, or to investors. There were no perfect answers, only judgments made inside narrow windows of time.

Leonard Schleifer of Regeneron captured the truth clearly: “In biotech, you have to be an optimist who is also terrified. You are betting on biology, and biology does not read your business plan.” The hardest working people in the industry are often working with the least certainty. That is why trials fail so often: not because leaders are careless, but because the information landscape is fractured and biology does not cooperate with human timelines.

Four forces make this work uniquely difficult. Biology itself humbles everyone who touches it. It is nonlinear, adaptive, and full of emergent behavior. Two cancer biologists may not even share the same conceptual language. Knowledge diverges faster than it converges. A CEO making a “scientific decision” is triangulating across dozens of narrow experts, each holding a fragment of an impossibly large system.

Then there is the market, where uncertainty meets capital. A biotech CEO must understand risk pricing as intimately as pathway biology. Every program competes not just with disease but with time, capital, and psychology. The information is probabilistic, opaque, and delayed. Even the best investors cannot truly know which drug will succeed. Volatility is extreme because the underlying uncertainty is extreme.

Operations add another dimension of difficulty. Clinical trials are miracles of coordination: dozens of sites, hundreds of patients, thousands of variables, all needing to align without error. Enrollment stalls, protocols are misinterpreted, manufacturing deviates, supply chains misfire. John Maraganore of Alnylam summed it up simply: “In biotech, the science will humble you, but operations will kill you.”

The regulatory pathway is a labyrinth of precedent and context. Success depends not only on what the data shows but on how it is framed, justified, and aligned with ethical standards. Every comparator, every endpoint, every phrase in a briefing document carries weight. You are not selling a product. You are earning trust.

This is why biotech is not “just another industry with normal challenges.” The layers interact. Biology, business, operations, and regulation each have their own data, incentives, and epistemology. When combined, the system becomes so complex it approaches chaos. No single human can hold all the relevant information at once. The bandwidth of the mind is too narrow. Expertise is siloed. The signal is buried in noise.

This is why over 90 percent of drug programs entering clinical trials fail. Not because teams are incompetent, but because the information is incomplete, the variables uncontrollable, and the underlying biology full of surprises. Even so, the stakes remain enormous. A successful trial creates billions in value and saves lives.

Both the influenza and covid pandemics revealed the same truth. Clinical trials are difficult because nature is difficult, and pandemics do not wait for us to get organized. COVID succeeded because decades of preparatory work already existed. The tools were built before the crisis.

The Role of AI

The question now is whether artificial intelligence can help. It can, but only if we are honest about what the problem really is. The main bottleneck in health is no longer discovery. We already have AI that can design proteins, search chemical space, and propose new molecules in silico. The bottleneck is delivery. Most drugs do not die in the lab. They die in the fragile, noisy, expensive process of proving that they work in humans.

This is where AI matters. At its best, AI is a compression engine for complexity. It can read every trial report, regulatory document, safety signal, and mechanism paper in minutes. It can model how a trial is likely to behave before anyone enrolls the first patient. It can stress test different designs, simulate enrollment patterns, and surface which subgroups are likely to respond or fail. It can coordinate thousands of decisions across endpoints, cohorts, comparators, safety monitoring, and site selection, decisions that today exist in separate slide decks and separate minds.

But AI by itself is not the point. A large model sitting on top of a broken process does not fix the process. If anything, it can make overconfidence cheaper. What we need is not AI as a toy or a chatbot. We need AI as infrastructure. An intelligence layer that makes the entire clinical system learn, not just answer questions one at a time.

Modern healthcare spends over eleven trillion dollars a year. Drug development alone burns more than one hundred billion dollars annually on clinical trials, with fewer than eight percent of programs reaching patients. That is not just biology being hard. That is a system that does not compound its own experience. Each company repeatedly pays the full price of uncertainty, rather than drawing down on a shared base of learning. AI is suited to change that. It is good at one thing humans are bad at: integrating massive amounts of messy information and extracting structure from it.

The next great company in health will not be another drugmaker or insurer. It will be the platform that makes medicine learn from its own experiments. This will also be the next great tool in our toolkit to address the next pandemic.

Clinical Trial Intelligence

CTI is needed now, even if the next major pandemic is not occurring now.

Clinical trial intelligence would sit across the life of a trial program. Before a trial starts, it would simulate different designs and flag those that are likely to fail based on similar programs in the past. It would identify which endpoints are fragile, which inclusion criteria are unrealistic, and where bias is likely to creep in. During a trial, it would track enrollment patterns, site performance, and data drift in real time, and suggest adjustments before problems become fatal. After a trial reads out, it would absorb the outcome and recalibrate its beliefs, so that every failure increases the probability of future success rather than simply erasing value.

This is not science fiction. The data already exists. Regulators publish decisions. Companies publish protocols, labels, and partial data. Trial registries track dates, locations, designs, and outcomes. Investors move billions on their interpretations. Right now, this information sits in PDFs, static websites, and human memory. Clinical trial intelligence is about pulling that entire universe into a single learning system that is trained on what actually happened, not just what was planned.

The impact of even modest improvement is enormous. For pandemics, this matters even more. When a new virus arrives, we will not have time to rediscover how to run good trials under pressure. We will need systems that have already learned from thousands of past trials in vaccines, antivirals, and immune modulation, ready to propose designs, anticipate pitfalls, and guide decisions in days, not months. The same platform that makes oncology and autoimmune trials smarter in peacetime is the one that will let us move at the speed a pandemic demands.

So the role of AI here is simple and ambitious. Make clinical trials self improving. Turn every protocol, success, and failure into fuel for the next decision. Give the hardest working people in the hardest industry a system that finally matches the scale and complexity of the problems they are trying to solve. When that exists, drug development will stop feeling like a casino and start feeling like progress.

Build Now

We cannot wait to build this when the next pandemic begins. The Gates Foundation and BioNTech did not know COVID was coming, but the mRNA platform built for HIV was ready when needed. That preparedness looked like foresight, but it was simply discipline. We know far more today. We know another pandemic will come. Climate change collapses the distance between humans and animals. Zoonotic spillover accelerates. Global travel speeds up transmission. The next pandemic could be fifty years away, ten years away, or already incubating somewhere we are not watching.

Clinical trial intelligence takes years to build and refine. Its value begins immediately, not only during a crisis. Oncology, immunology, infectious disease, and rare disease trials all benefit today. The same system that strengthens current development is the one that will pivot when the next outbreak arrives. The way mRNA pivoted. Only faster.

We build it now. Not during the crisis. Not after. Now, while we still have the luxury of preparation. Another pandemic is coming, and this time we already know it.

***
World Health Organization (WHO), “A World at Risk” (Global Preparedness Monitoring Board, 2019).

National Academies of Sciences, Engineering, and Medicine (NASEM), “Biodefense in the Age of Synthetic Biology” (2018).

Nature Editorial, “Artificial intelligence and the potential for biological misuse” (Nature, 2023).

The hardest-working people should work on the hardest problems.

For me, that problem set is in health. Right now, some of the hardest problems in health are in healthcare and drug development.

Few human endeavors combine as much uncertainty, intellect, and consequence as bringing a new therapy to life. It is science, business, operations, and regulation, all colliding in real time, under extreme pressure, with billions of dollars and human lives in the balance.

I learned this firsthand running Healis Therapeutics, where I became the youngest cofounder and leader of a clinical-stage neuroscience therapeutics company in the world. I had to make decisions that even seasoned executives lose sleep over. Every day, I faced questions that had no single right answer: Which indication should we prioritize? How will regulators interpret our trial design? What data matters most to investors, or to patients?

Each decision was a bet made on imperfect information. Like Leonard Schleifer, Founder & CEO of Regeneron, once said: “In biotech, you have to be an optimist who’s also terrified. You’re betting on biology, and biology doesn’t read your business plan.”

To develop a drug is to live in uncertainty, where the data is incomplete, the incentives misaligned, and the timeline unforgiving.

The Four Dimensions of Difficulty

Drug development sits at the intersection of four domains, each complex on its own, but nearly impossible when combined.

1) Biology: she who humbles everyone who touches it
It is the most complex system in the known universe, nonlinear, adaptive, and full of emergent behavior. No two PhDs in biology share the same foundation; even two cancer biologists might struggle to talk meaningfully about each other’s work. One studies DNA repair, another immune signaling, and their worlds barely overlap.

That’s the paradox of modern biology: the deeper you go, the narrower your tunnel becomes. Knowledge diverges faster than it converges. Every discovery spawns ten subfields, and every model works only under certain assumptions. So when a CEO makes a “scientific” decision, they are triangulating across dozens of experts, each seeing a fragment of a reality too large for any single human mind.

2) The market, where uncertainty meets capital
A biotech CEO must understand not just biology, but risk pricing. Every molecule competes not only with disease but with time, capital, and investor psychology. You must decide which indications are viable, which patents are defensible, and which milestones can sustain valuation.

Biotech markets are inherently inefficient because the underlying information is probabilistic and opaque. Even the most sophisticated investors can’t truly know which drug will succeed. That’s why volatility is extreme, and fortunes are made or lost overnight.

Vas Narasimhan, CEO of Novartis, best summed it up: “You can have the best science in the world, but without a sustainable business model, it’s just an academic exercise.”

3) Clinical Trial Operations: a miracle of coordination
Dozens of sites, hundreds of patients, thousands of variables, and everything must align perfectly. Enrollment stalls, data pipelines glitch, CROs misinterpret protocols, or a single manufacturing deviation triggers a cascade of delays. The smallest operational error can destroy months of progress and millions in value.
“In biotech, the science will humble you, but operations will kill you” in the words of John Maraganore, Founding CEO of Alnylam Pharmaceuticals

4) Regulatory: the labyrinth

The FDA and EMA are not adversaries, they’re stewards of trust, but they speak a language of precedent and proof. Success depends not just on what your data shows, but on how it is framed, contextualized, and justified. The path to approval is paved with nuance: every phrase in a briefing document, every comparator chosen, every safety flag raised.

Is biotech really exceptionally difficult?
But doesn’t every CEO face operational, market, and regulatory challenges? Biotech is no exception, right? But in this field, the complexity runs far deeper. The science itself poses some of the most difficult decisions: navigating the intricate nature of biology, managing the operational challenges of conducting precise clinical trials, and sustaining the immense financial investment required for research and development. On top of that, biotech operates within one of the most tightly regulated environments, from the FDA to market approval, involving countless stakeholders. These include governments, insurers, healthcare providers, and more. All of whom must align before the ultimate customer, the patient, can benefit. Unlike most industries, you cannot simply launch a product based on consumer demand or willingness to pay. The path to market in biotech is a winding one, with few clear roadmaps and limited opportunities to learn from past examples.

Living in a World of Uncertainty

When you combine these four layers, scientific, business, operational, and regulatory, you get a system so complex it borders on chaotic. Each layer has its own data, incentives, and epistemology. No one, not the scientists, not the investors, not the regulators, can see the entire picture.

That’s why biotech is still, in many ways, an era of intelligent gambling. We make our best bets with partial data and hope that biology cooperates. The result is staggering inefficiency. Over 90% of drugs that enter clinical trials fail. Biotech stock prices swing by 50–80% on single readouts. And even “experts” often disagree on which programs will succeed or why.

This isn’t because people are careless, it’s because the information space is fractured. The human mind can’t integrate the full spectrum of molecular, clinical, operational, and economic signals simultaneously. Every decision is filtered through the narrow bandwidth of human cognition and the silos of expertise.

The outcome is a landscape defined by uncertainty and inefficiency, one where insight is scattered, coordination is imperfect, and progress comes at enormous cost.

“If you can’t live with ambiguity, you can’t live in biotech.”
- Bob Langer, Co-founder of Moderna

When It Works, and When It Doesn’t

When it goes well, it’s a masterpiece of coordination and courage.

Look at BeOne’s success in leukemia, a precise scientific bet, executed with operational discipline, aligned with regulators and investors alike.

Billions in value created, and many lives saved. A cancer breakthrough to marvel at.

But when it goes wrong, it’s devastating.

Think of Neumora’s failed Phase 3 in depression, a single trial readout that erased $3 billion in market cap overnight. The same molecule, the same scientists, the same commitment, but biology said no.

This is exactly why I believe running a biotech is the hardest job in the world.

You must be fluent in four languages, scientific, commercial, operational, and regulatory, and translate between them every day. You must hold conviction under uncertainty and humility under success.

The Opportunity Hidden in the Chaos

But amid this chaos lies opportunity, because chaos has structure, if you can find it.

The inefficiency of biotech markets isn’t just a risk; it’s a signal. The noise hides patterns, correlations between trial design, mechanism, regulatory precedent, and real-world outcomes. The data is there; it’s just fragmented, unstructured, and uncalibrated.

What if we could make sense of it?

What if we could collect, integrate, and annotate every piece of trial data, mechanistic knowledge, regulatory precedent, and market outcome, and train on it?

What if we could develop an intelligence that doesn’t just memorize, but reasons, across domains, across silos, an intelligence that can see connections no single human can?

A Clinical Trial Intelligence:

1. Trained on the world’s biomedical, operational, and market data,

2. Calibrated to the real-world success and failure of clinical trials,

3. Capable of making cross-domain decisions like the collective mind of 10 million experts.

An agent that thinks like a scientist, operates like an executive, and reasons like a regulator.

A synthetic super life sciences executive, one that can simulate, predict, and optimize across the hardest problems in the world.

Because what this industry needs isn’t more data.

It needs understanding. And understanding comes from learning.

The Future of Biotech Leadership

The future of drug development will belong to those who can integrate, not just specialize. The leaders of tomorrow will partner with systems that think alongside them, not replacing human expertise, but amplifying it.

If we can build an intelligence that bridges biology, business, operations, and regulation, that makes sense of uncertainty and discovers the hidden structure in the chaos, then perhaps, for the first time, drug development won’t feel like gambling.

It will feel like progress.

And maybe then, the hardest-working people in the world will finally have the tools they deserve, to move the hardest problems in the world, faster and further than ever before.

Sources:

IQVIA Institute. The Global Use of Medicines: Outlook to 2028 (Jan 2024) https://www.iqvia.com/insights/the-iqvia-institute/reports-and-publications/reports/the-global-use-of-medicines-2024-outlook-to-2028 IQVIA+2IQVIA+2

Biotechnology Innovation Organization (BIO). Clinical Development Success Rates 2011–2020 https://www.bio.org/clinical-development-success-rates-and-contributing-factors-2011-2020 BIO+1

ScienceDirect / Elsevier. Benchmarking R&D Success Rates of Leading Pharmaceutical Companies https://www.sciencedirect.com/science/article/pii/S1359644625000042 ScienceDirect+1

Alzheimer’s Disease International (ADI). World Alzheimer Report 2022 https://www.alzint.org/resource/world-alzheimer-report-2022/ Alzheimer’s Disease International+1

JAMA Oncology. Estimates and Projections of the Global Economic Cost of 29 Cancers: 2020–2050 https://jamanetwork.com/journals/jamaoncology/fullarticle/2801798 JAMA Network+1

American Heart Association (AHA). Heart Disease and Stroke Statistics 2024 https://www.heart.org/-/media/PHD-Files-2/Science-News/2/2024-Heart-and-Stroke-Stat-Update/2024-Statistics-At-A-Glance-final_2024.pdf?hash=D0A208F50F8591AEED0E31BE77265505&sc_lang=en www.heart.org+1

United Nations. World Population Prospects 2024 https://population.un.org/wpp/ population.un.org+1

Colonel Eli Lilly, Indianapolis (1893)

The birth of biotech
The first great wave of modern biotech came in the 1980s, when companies like Genentech and Amgen first turned biology into engineering. Think recombinant DNA, monoclonal antibodies, genetically modified cells. The second wave came in the 2000s, when genomics, computation, and venture capital collided: sequencing costs fell, startups multiplied, and biotech turned moonshot science into moonshot startups. Yet today, that biotechnology world is showing its age. Drug approvals have slowed, the cost of development has skyrocketed, and the system built by scientists and financiers alike has begun to strain under its own complexity.

TechBio, the California Roll we have all been waiting for
The next renaissance will not come from biotech as we know it, but from TechBio and AI for Bio. The merging of artificial intelligence and life sciences.

TechBio will be like the California roll: it took Californians to flip sushi inside out and create something that the Japanese themselves had not dared to do. In the same way, it will take technologists to reimagine how biology is built; to break old taboos, automate what was once artisanal, and make medicine scalable, flipping Biotech into TechBio.

At the time that the California Roll was invented (1967), turning sushi “inside out” was considered a taboo in Japan, breaking the tradition hat had defined centuries of sushi.

Lessons from software
Twenty years ago, software looked like biotech does now. It was expensive, slow, and controlled by incumbents. The early software world was MBA-led, not founder-led; investors incubated their own companies, hired their own CEOs, and treated devs as replaceable labor. Then came open-source software, cloud infrastructure, and global distribution. Two cofounders in a dorm room could suddenly reach the world. The barrier wasn’t capital; it was creativity. Once cost fell and access widened, founders took control. Even in regulated industries like finance, entrepreneurs broke through. PayPal and later Robinhood proved that even in highly regulated domains like finance, technology could advance.

So where is the PayPal of pharma? Where is the founder-led company that rewires how medicine is developed and tested? It doesn’t exist yet, because the cost of biotech entrepreneurship hasn’t dropped to levels of software yet. Biotech remains in its pre-cloud era: pharma-led, capital-heavy, and risk-averse. Most “startups” are still conceived in venture studios, built by hired executives, and run under investor control. That model was somewhat justified when trials cost millions. It doesn’t work now. Trials now take 10 years, $1 billion, and succeed less than 10% of the time to produce a new therapy.

The acceleration
The change biotech needs is the same one that transformed software: the collapse of cost and complexity. As long as each trial, experiment, and regulatory cycle becomes more expensive each year, no founder can move fast enough to challenge the system. The revolution will come from making the process itself cheaper, faster, and smarter; turning biology into an engineering problem, not a bureaucratic one.

The first signs of this are already visible. Biology is becoming computational. At Profluent, new proteins are designed and tested entirely in silico. Genesis Therapeutics uses graph neural networks to model molecular behavior. Xaira Therapeutics generates drug candidates nature never imagined. In the clinical layer, Formation Bio and Reify Health are rethinking trials as intelligent, adaptive systems. These companies are early prototypes of TechBio: small, technical, founder-led.

Clinical trials hold the answer
Today, the bottleneck remains. Founders can now create brilliant molecules in weeks. Yet it still takes years to prove them. The clinical trials phase is where most innovation dies. Everyone is betting on AI for drug discovery, and that bet is simple: that intelligence can compress the search space for new molecules. But good molecules, and even great ones, still fail in clinical trials. They fail not because they are wrong, but because the trial is. Design flaws, patient selection, operational noise, and unobserved biases destroy billions of dollars of potential. What is missing is another layer of intelligence. An intelligence not focused on discovery only, but on delivery; not on the molecule, but on the path that brings it to patients.

Software became founder-led because it became easy to build and test ideas. Biotech will become founder-led when it becomes easy to run and validate them. Once experimentation is cheap enough and trials are intelligent enough, small teams will be able to build, test, and iterate in medicine as fast as they do in code. This is the task of the PayPal of pharma, the company that rewires the system from the bottom up by making it ten times easier to build within it.

Intelligence for delivery, not just discovery
What if a company was built entirely focused on developing Clinical Trial Intelligence?Building an AI layer that increases the success rate of clinical trials and helps decision-makers see what’s coming before it happens? An intelligent clinical trial can predicts trial outcomes, identify failure risk early, and enable sponsors to intervene in trial design and operations in real time. By connecting biological insight with operational foresight, Clinical Trial Intelligence would help pharmaceutical companies develop drugs more intelligently, efficiently, and successfully.

The cost curve will bend. It always does. Computation has already rewritten discovery. Automation is beginning to rewrite development. A new company will emerge to rewrite clinical trial validation, which is the hardest and most expensive phase of all. The next generation of founder-led biotech will not just make better drugs; it will make smarter ways to prove they work. When that happens, biology will finally catch up to software; not by imitating it, but by becoming it.

In 1998, the future of finance looked less like Bank of America in New York City and more like PayPal in Palo Alto. In 2025, intelligent pharma looks less like Eli Lilly in Indianapolis and more like a young startup in Mountain View.

Disclaimer:
This essay is an educational and speculative exploration of how a company might be founded in today’s technological landscape. All references to Eli Lilly are for historical and illustrative purposes only. It is not affiliated with, endorsed by, or representative of Eli Lilly and Company or any of its employees, subsidiaries, or affiliates. Or any company mentioned in the article.

In 1900, the average American lived just 47 years. By 1950, it was 68. By 2000, nearly 78. (CDC)

That’s a 30-year leap in a single century. One of the greatest health transformations in human history. Vaccines and antibiotics wiped out infectious disease. Clean water and sanitation became universal. America learned how to deliver babies safely, manage chronic conditions, and nourish entire populations.

For a hundred years, the health graph of America pointed relentlessly upward. Then, around 2000, it didn’t.

The Plateau

From 2000 to now, U.S. life expectancy grew only by a year, from 78 to 79. It dipped during COVID and still haven’t recovered. (CDC). For example, Between 2010 and 2019, the U.S. gained just 0.1 years of life expectancy, while peer nations gained more than a full year. (Health System Tracker)

It’s not because America stopped caring. America spends more on healthcare than any country in the world. Yet the return on that investment, such as longer, healthier lives, has stagnated. The curve that defined a century of progress has gone flat.

Two Americas

Of course, life expectancy isn’t everything. It tells us how long we live, not how well. Living to 85 but spending 20 of those years fighting chronic disease isn’t success. Quality of life matters.

And the averages hide another harsh truth. In some U.S. counties, life expectancy still reaches the 80s, while in others it’s closer to 60. (CBS News) The gap between zip codes can mean 10 to 20 years of life. (Hamilton Project)

That’s not just a plateau, but a divergence; two health trajectories inside one country.

The Innovation Slowdown

If the 20th century was about conquering infection, the 21st was supposed to be about conquering everything else that is non-infectious: cancer, Alzheimer’s, autoimmune disease. But the rate of new medical breakthroughs has slowed. Clinical trials are more expensive, slower, and riskier than ever. The industry has grown cautious, often chasing incremental gains instead of transformative cures.

Big bets are made on AI to reinvent drug discovery, which is promising. Yet discovery is only half the equation. The other half is getting those discoveries through clinical trials and into patients’ hands.

Right now, that system is broken. Trials fail too often, take too long, and cost too much. Not always because the science is weak, but rather because the data is. We need a new layer of intelligence for clinical trials. A way to make clinical trials learn faster from themselves. That’s how medicine will start compounding again.

The Real Health Equation

Health has many dimensions. Some we can change. Here is an overview:

Health = Lifestyle + Healthcare

1. Lifestyle = Fitness + Nutrition + Sleep + Stress

2. Healthcare = Access * Quality
   

Access is about policy: who gets care, when, where, at what price, and who pays for it? It’s shaped by coverage, cost, geography, and demography. Quality, on the other hand, is about how well and how safely treatments work; whether it’s a drug, surgery, or device.

Most people believe access is the main issue in American care. A lot of hardworking young people I know focus their careers on health policy and equity. This matters, but access is ultimately a political fight and in today’s divided climate, progress is tough. Quality is where we have more room to make real change. It’s the part that drives innovation. Lately, it’s stalled. The biggest issue in healthcare right now isn’t just access alone, but quality. That is because access and quality share the same root solution: a more competitive and dynamic pharmaceutical industry. More pharma companies, more innovators, more researchers becoming founders, and more overlap between pharma, biotech, and techbio startups would mean more treatments to market, more competition, and more choices. That competition would naturally drive prices down and make care more available as we wait for laws or policies to change. It’s the market’s way of creating progress now. It gives us builders agency, now, to build better healthcare, as we wait for political changes that are far from guaranteed.

The Next Health Curve

From 1900 to 2000, America built the public health systems that lifted an entire nation. From 2000 to now, America has been coasting on their legacy.

The next leap won’t come from more hospitals or more coverage alone. It will come from making the system itself smarter. Clinical trials that learn, data that connects, therapies that reach patients more intelligently.

We don’t just need more drugs. We need a better way to prove they work. That’s how we get the curve rising again. In the next article In the next article, I’ll explore what Intelligent Pharma really means. What does the future of clinical trials look like? How will the next generation of pharma, biotech, and care systems transform the way we develop and deliver treatments? A healthy future awaits us and we have the agency to deliver it sooner rather than later.

Watch: “What Happened to Our Health”:

Every year, pharma spends over $120 billion running 30,000+ clinical trials. But less than 8% of drugs succeed through clinical trials to make it to the doctor’s office or pharmacy. Each failure means lost time, burned capital, and another therapy that never reaches patients. The problem isn’t always the science; it’s often the data. Trial data remains fragmented and inconsistent, with insufficient granularity to support meaningful learning. (Source: WHO; Nature Reviews Drug Discovery, 2025).

When I first saw those numbers, it hit me how much potential is being wasted, not because the molecules are bad, but because we’re flying blind even in the later critical stages of drug development.

We’re building the first Trial Intelligence platform. We’re betting on AI designed not to discover drugs, but to increase the success rate of clinical trials. Even great molecules fail in clinical trials. Trial Intelligence is the layer necessary to get the next cancer treatments, or that long sought Alzheimer’s treatment, over the finish line and to the patients who need them.

We have spent years in both AI, pharma, and social good. Between us, we’ve gained experience building AI that solves real-world problems and taking drugs from concept to clinic to market. I’ve lived the reality of running clinical programs and seeing how fragile those systems are in practice. Fei brings deep research experience at Tsinghua and Caltech, and technical expertise in pioneering systems that calibrate AI to real world needs.

Here’s the crazy part: improving trial success rates by just 1% could unlock over a billion dollars in value every year. (Source: WHO; Nature Reviews Drug Discovery, 2025).

That’s the future we’re calibrating toward.

What the world could look like
Today, pharma already pours over $120 billion a year into clinical trials, yet most investors stay on the sidelines, wary of the high failure rates. The risk is so great that even promising drugs often die before they get a chance to prove themselves. If we could increase the success rate of late stage clinical trials, that dynamic would flip. Over a trillion dollars in new capital will flow into the space, funding innovation and novel therapies that are currently too uncertain to touch.

That would mean more shots on goal: more cancer drugs reaching approval, more Alzheimer’s and autoimmune therapies crossing the finish line, more rare disease programs finally getting a real chance. But beyond the numbers, it would mean less human suffering, patients living longer, healthier lives, families spared from loss, and entire communities lifted by better access to care. Healthcare systems would learn faster. Progress would accelerate. Humanity would get a health upgrade.