THE FUTURE OF PHARMA IS HERE

Agents are earning the right to operate in the $1.7 trillion pharmaceutical industry

It is now clear that one day a $100B pharma company can be run by only 7 people and their agents. This full implementation will take a decade. But we are seeing the beginning. 2026 is the year pharma agents start making deeper headway in real workflows.

The future of pharma is here. Not because of a breakthrough model. Intelligence is here. But because the $1.7 trillion pharma industry has hit the real constraint: deployment. In a safety-sensitive, regulated domain, intelligence isn’t enough. Agents must earn the right to operate. And that right is set by the workflow itself.

Agents 101
For an agent to automate, a workflow must be:
- defined
- data rich
- deployable.

Defined means the inputs are clear, the steps are repeatable, and the outputs can be judged against explicit criteria. Data rich means the workflow leaves behind enough signal to train on, audit, and improve. Deployable means the workflow can be executed safely with traceability and governance, first with close oversight, and later with limited oversight.

In pharma, “deployable” is not a convenience label. If you think pharma agents will win by intelligence alone, you’re looking in the wrong place. They’ll win by earning the right to operate inside real workflows and become truly deployable.

Supervision
Yes, one day a $100B pharma company will be run by only 7 people and their agents. Pharma is not adopting agents overnight. Agents begin in tightly controlled settings where workflow actions are logged and outputs reviewed, then progress to co-pilots under close supervision. Only after proving reliability within clear rules do they earn limited autonomy. And when workflows are less defined, less data rich, or less deployable, that path gets longer, with fewer signals to learn from and more ways to fail silently.

The stakes explain the caution. Industry scale analyses show that the overall likelihood of approval from Phase I across 2011 to 2020 was about 7.9%. It takes 10 years, a billion dollars, and a less than 10% chance to succeed. It can often be more than 10 years, a far lower likelihood, and $3B+ invested. If your workflow is a machine that burns a decade and billions of dollars, “almost correct” is not an acceptable operating point.

The first Pharma Agent
Once you accept that agents are a workflow problem, the natural next question is: where do agents deploy first?

I think the cleanest way to see this is through the “Magnificent Seven” workflows of pharma leadership: the CEO function, the CSO function, the CBO function, Clinical and Medical, Regulatory, CMC, and IP. These are the seven engines that shape what gets built, tested, approved, manufactured, protected, and financed.

They are not equally ready for agents.

Some functions are extremely data rich but poorly defined. That is science. CSO workflows sit on vast data, and AI for science will keep advancing, but science is messy. Processes are not standardized, novelty is the goal, and novelty resists automation. A workflow can be data rich and still be hard to deploy into if the steps are unclear and outputs cannot be audited.

Other functions are more defined but constrained by safety and regulation. Regulatory and CMC work produce structured artifacts with clear quality bars, but error tolerance is low and consequences are immediate. IP is defined but adversarial and jurisdictional. Clinical and Medical is data rich and increasingly instrumented, but the deployment surface is large and safety implications are direct.

CEO work is an interesting case. It can be highly data rich at the portfolio level, but it is often the least defined as a single workflow. Strategy is a set of choices, not a form.

Deployability made simple
If you rank the Magnificent Seven by how agentifiable they are right now, the key dimension is deployability under safety constraints. “Deployable” in pharma means: can an agent operate inside an oversight regime that produces auditable artifacts, avoids uncontrolled actions, and stands up to the scrutiny that regulated industries demand.

On that ranking, there is one clear leader.

The CBO function.
BO workflows are the most defined, data rich, and deployable because they already operate as decision support with clear human signoff. Deals close through accountability, not models, and the standardization of BD artifacts makes these workflows well suited for oversight-first agents.

CBO is also where document volume and fragmentation are most visible. Programs generate large, scattered corpora, and due diligence exists because no one person can see the full picture and errors here compound into years of wasted capital.

Two CBO workflows matter most for agents: search and evaluation, and due diligence. Search and evaluation form the front door, continuously scanning and triaging opportunities. They are repetitive and criteria driven, which allows agents to learn what matters, flag risks, and produce consistent evaluation memos.

The first business agent will be in pharma due diligence.

Pharma DD
Due diligence is the engine room where agents become inevitable because the work is structured but overwhelming. The question is not whether an agent can write a memo, but whether it can produce auditable, calibrated, evidence grounded decision support.

Two sub workflows sit at the core of BD diligence. Probability of success modeling turns scattered evidence into explicit phase transition estimates. Risk adjusted valuation, often rNPV, converts probability, timelines, and economics into models that can be debated and tested. These are standard human artifacts, which is why the workflows are deployable. Agents do not replace judgment. They improve the speed and quality of the inputs judgment relies on.

This is exactly where we start at Panoptic Bio.

Clinical Trial Outcome Prediction
We built a clinical trial outcome prediction system that predicts the probability of phase transition, including Phase I to II, Phase II to III, and Phase III to approval. On the TOP/HINT benchmark, using the published evaluation protocol, we reported an ROC AUC of 0.76, and we evaluate both discrimination and calibration, since calibrated probabilities are what diligence actually needs. We also reported that our evaluation was run on a substantially larger corpus, spanning 11,600 historical clinical trials and 560,398 unique documents across scientific, regulatory, commercial, and operational sources. That scale matters because it makes it possible to benchmark a program against real precedents, not just produce a plausible narrative.

The deployment philosophy matters as much as the model. We do not start by dropping agents directly into a pharma company’s internal BD workflow and tooling. We start with a safe place to deploy, train, and observe.

That place is the Trial Terminal.

Trial Terminal: A Safe Space to Test Pharma Agents

Trial Terminal gives users and their agents a controlled environment to retrieve evidence and produce diligence work products with full traceability. It is designed for due diligence, portfolio analysis, and go or no go decisions, making safe deployment possible in a regulated domain. Agents do not get unrestricted system access. They produce artifacts that humans review and correct, and oversight becomes the training signal.

This may look slower than instant autonomy, but it is how trust is earned in regulated industries.

From the outside, this can look slower than the dream of instant autonomy. In reality, it is the only way to earn trust in a regulated industry.

Timeline
This brings me to the timeline question, and to a conversation I had with a very prominent VC in Silicon Valley. I told him that the full process of agentifying pharma will take 10 to 15 years. He worried that his four year investment horizon would not match that. I replied, half joking, “Ok, see you in 6 to 11 years then?”

But I do not think that is the right way to think about it.

“Ten to fifteen years” describes the full end state. It does not describe when value begins.

Waymo
Waymo is a useful analogy. The Google Self Driving Car Project began in early 2009. Waymo opened its fully driverless service to the public in Phoenix on October 8, 2020. That is an eleven year arc to a headline milestone, but the point is not the number. The point is that value was created through milestones long before “full autonomy everywhere” was real.

Pharma’s Waymo Ride to Autonomy
Pharma agents will follow the same curve. The full transformation is a decade scale shift because workflows must become defined, data rich, and deployable across the organization, under safety and regulatory constraints. But significant value is realized much earlier, including this year, by agentifying the first workflow the right way. One workflow at a time. Bit by bit. With oversight first deployment, measurable performance, and continuous improvement.

That is the strategy. Start where the workflow is most defined, data rich, and deployable. That is CBO. Deploy agents into search, evaluation, and due diligence. Anchor them in probability of success modeling and rNPV analysis. Train them inside a safe environment like Trial Terminal where accountability is clear and learning is grounded in real review cycles.

Agents for Hire
One day, when more of pharma’s workflows are truly defined and instrumented, this becomes the future across the Magnificent Seven. Agents will expand from BD into regulatory intelligence, IP reasoning, clinical and medical operations, and even parts of portfolio strategy. CSO workflows will benefit enormously too, but they will arrive later in the deployment curve, because science is data rich but not consistently defined, and the messiness of novelty makes deployability harder.

Example of the safe deployment of pharma agents in the Trial Terminal

The year ahead
2026 is the year this starts to look real because the industry is moving from fascination to workflow integration. The path is not magical. It is disciplined. Defined workflows. Data rich signals. Deployable systems. Close oversight first, then limited oversight. That is how you build pharma agents that create real value for pharma teams, pharma investors, and ultimately patients.