LinkedIn Series: Five Questions Every Biotech Leadership Team Is Asking

Five Questions. FiveAnswers.

These are not vendor evaluation criteria. They are the questions that determine whether AI is going to make your company stronger or quietly add risk to programs you are already running close to the edge. When we set out to build Oktopi, we wrote down five architectural principles before we wrote aline of code. Not values. Not slogans. Architectural commitments. The things the platform must do, structurally, to support medicine development as it is actually done. Each of those principles is the answer to one of the five questions above.

‍

Question 1: AI that stands up to regulators and investors

The one I always get asked first.

How do we use AI in a way that stands up to regulators and investors?

This is the question that decides whether AI becomes infrastructure for your company or aliability quietly accumulating on your balance sheet. If the platform youchoose cannot survive scrutiny from a regulator, a board, or a sophisticated investor in a diligence room, none of the productivity gains matter.

There is apractical test for whether a platform satisfies this principle. It is the testmost platforms quietly fail.

Would your CMO, CSO, and CDO all sign off on its outputs?

Demos pass when one person nods. Production passes when three accountable executives, each with different evidentiary standards, all accept the output. The CMO needs clinical defensibility. The CSO needs scientific rigor. The CDO needs development logic that holds up to a regulator and a board. A platform that produces output passable to one of them and shaky to the others is a productivity tool, not a development platform.

And there is a second question underneath the first. Where does your data live while the platform is producing those outputs? We do not train on your data is a contractual promise. Where the data physically lives is an architectural fact. Even the best intentioned vendor terms are subject to change, breach, or interpretation. The architecture is not.

We built Oktopi as the kind of platform three accountable executives can all sign off on. Every output traceable to evidence. Every claim provenanced. Every decision recorded as a human decision. And Oktopi itself is deployed inside your own cloud tenancy, in your region, scoped to the specific project. Your data does not leave that environment. Models do not learn from your data, by architecture, not by promise.

AI does not develop drugs. People develop drugs, supported by AI that earns their trust by being transparent about what it knows, what it does not, and how it reached the conclusion in front of them.

‍

Question 2: Teams thinking together on AI

The one every leadership team feels in their meetings.

How do we get our teams thinking together on AI, instead of in five different chat windows?

Here is a meeting I have sat through more than once. Three of my team walk in. Each has used a chat based AI separately to prepare. Each has a slightly different answer. The meeting becomes an argument about which AI output is right, instead of a creative discussion about the asset.

The platform that should have facilitated the conversation has actively fragmented it.

This is not a workflow problem solvable by sharing chat logs. It is architectural. Generic AI is built for one user, one prompt, one response. Even when teams use it, they use it in parallel, each in their own window, each generating their own output, each importing it back into a shared context the model never sees.

The decisions that matter in medicine development are not retrieved. They are made, in rooms, by teams. Target product profile. Modality choice. Indication strategy. Dose rationale. None of these is solved by one person typing into a chat window, however clever the model.

And there is a second part to this answer that matters just as much. The platform must anticipate, not wait. Generic AI rewards the prompter. If you already know what to ask, you get sharper answers. If you do not, you get confident sounding, plausibly structured, often subtly incomplete answers, with no signal that you should have asked something else entirely. The hardest part of running a program is not answering questions. It is knowing which questions you have not asked yet.

What lean biotech needs is the opposite shape. Interactive surfaces where the team explores together, with shared assumptions visible to everyone in the room, structured divergence and convergence, and outputs the team owns rather than chat logs no one revisits. And a platform that surfaces the next question before you have formed it, flags the blindspot you did not know you had, and puts the right exploration in front of you when you arrive at the right stage gate.

We built this into Oktopi as the default state of the platform. The team thinks. Oktopi structures, anticipates, and surfaces. It is a different category of product than a chat window.

‍

Question 3: AI that understands our assets and our team

The one leadership teams ask when generic AI keeps giving them generic answers.

How do we get AI that actually understands our assets and our team?

Try this. Open whichever AI tool your company uses most. Ask it whether your lead asset is ready to advance to the next gate. What you will get is a plausible sounding paragraph. What you will not get is an answer grounded in your asset’s actual stage, your governance framework’s actual evidence requirements, or your development committee’s actual decision history.

The model does not know any of those things. It cannot. They were never built into its architecture.

A medicine development program is not a corpus of documents to be queried. It is a sequenced pipeline of decisions, each with prerequisites, each with evidentiary standards, each with named accountability. From discovery to filing, every meaningful question, progress to first in human, go or no go for proof of concept, ready for end of Phase 2, sits inside a governance framework that determines what counts as an acceptable answer. The structure has to be in the platform, not in the prompt.

And there is a second part to this question that lean biotech feels every day. The biotech operating model is lean by design. Your clinical lead also covers medical affairs. Your CMC head also covers quality and supply. Your CSO also fronts BD conversations. Your CEO is the de facto Chief Development Officer until well after Series B. This is not a failure of resourcing. It is the operating model of every biotech under a few hundred people.

Multi hat operators are not novices. They are deeply expert in their primary domain and competently navigating two or three more, while being held accountable for decisions across all of them. They ask reasonable questions in their secondary functions. A generic AI returns reasonable answers. The depth gap, the three follow up questions a single function specialist would have asked next, the adjacent function implications a regulatory or commercial lead would have flagged, is invisible to both the user and the model.

Oktopi was built around stage gate governance from day one. The platform behaves differently depending on stage, because it knows. And it calibrates to the function the user is operating in right now, not to their job title. The clinical lead asking a CMC question gets the inquiry depth a CMC head would apply.

Oktopi calibrates to the hat, not the head wearing it.

‍

Question 4: Fit for purpose help when resources are constrained

The one leadership teams ask when they realize enterprise AI was not built for them.

We are resource and capability constrained. How do we get fit for purpose help?

Most enterprise software in pharma is built for big pharma. That sounds obvious, but it has consequences few biotechs think through until they are six months into a contract.

Enterprise platforms come with implementation timelines measured in quarters, professional services bills measured in millions, change management programs that require dedicated headcount, and complexity that assumes the buying organization has IT, ops, and training teams to absorb it. They are built for organizations with the resources to deploy them.

Lean biotech does not have those resources. A small team with finite budget cannot afford a six month implementation, cannot dedicate headcount to running the platform, and cannot wait through change management cycles before getting value. They need infrastructure they can stand up quickly, run lightly, and adapt as the company grows.

This is not a feature problem. It is a fit for purpose problem. An enterprise platform that requires enterprise resources to deploy is, for a lean biotech, the wrong category of product, regardless of how good the features are.

We built Oktopi as the opposite of that. Light to deploy, agile to use, and designed for a team that does not have an IT department. The platform stands up fast, scales as you grow, and does not require anyone’s full time job to keep running.

This is what built for lean biotech actually means in practice. Not just designed for biotech use cases. Designed to be runnable by the small teams that biotech actually consists of.

‍

Question 5: Teams more capable, not more dependent

The closing question in this series.

How do we make our teams more capable, not more dependent on the tool?

This is the question most vendors hope you do not ask. It comes in two halves, and both matter.

The first half. AI cannot resolve every question a development program generates. The hard ones, novel modality positioning, contested regulatory strategy, complex CMC and clinical interfaces, judgment calls under uncertainty, require named human experts with regulated industry experience and accountability for their advice. A platform that pretends otherwise creates a different kind of risk than the one it claims to solve.

The honest test for any AI tool you are evaluating: when the model hits the edge of what it can safely resolve, what happens next? If the answer is the model produces its best guess and you decide whether to trust it, the platform has no operating model for the questions that matter most.

Lean biotech does not need a chat window that confidently answers questions it cannot. It needs a structured path from the AI to a human expert, inside the same platform, working from the same program context, with the output captured back into the company’s institutional memory rather than lost in an email thread or a slide deck.

The second half. Generic AI rewards the prompter. The expert who already knows what to ask gets sharper answers, faster synthesis, better outputs. The user who does not know what to ask gets fluent, plausible, often subtly incomplete answers, with no signal that they should have asked something else entirely. Both walk away from the same model with the same fluent text. Only the expert can tell which one is right.

This is the failure of generic AI that bothers me most. It widens the gap between novice and expert at exactly the moment when biotech needs that gap to close.

We built Oktopi to close it. Integrated expert access, where the path from the AI surfaces this question to a named expert provides an accountable answer is one click, and the answer becomes part of your program record. And a platform that models the inquiry pattern of an expert peer, surfaces the questions a senior practitioner would ask alongside the one in front of you, and teaches by example. Over time, your team becomes more capable across functions. Not more dependent on the platform.

That is the five. AI that stands up to regulators and investors. Teams thinking together on AI. AI that understands our assets and our team. Fit for purpose help when resources are constrained. Teams more capable, not more dependent.