Insights · AI in drug development
When AI Is Standard, Where Is the Edge in Drug Development?
Formation Bio's $600 million fund is a significant test of whether AI changes the economics of clinical development. The answer depends on separating three things that are easy to conflate: running trials faster, choosing better assets, and the financing structure that pays for both.
On October 28, Formation Bio announced a $600 million Drug Development Accelerator Fund to acquire or license 10 to 15 drug programs over three to five years.1,2 Each program sits in its own company and draws on Formation Bio's shared clinical, regulatory, manufacturing, and AI infrastructure.2 The company has said AI lets it run certain trials up to 50% faster.3
The obvious objection is the one many operators will raise first. A trial still needs patients, sites, investigators, drug supply, and follow-up, and AI does not change the biology or the evidence regulators require. So where does the advantage come from?
There are three candidate sources, and they are not equally strong. Trial execution and design are real gains, though narrower than the headline suggests. Asset selection carries more economic leverage but is the hardest to prove. And the most novel element of the announcement may be the capital structure rather than the AI.
Where AI moves the timeline
AI shortens a trial only when it speeds up the activity that controls the schedule. Faster protocol or regulatory drafting helps when that work is holding up study start. If enrollment is the constraint, the readout date does not move.
Recruitment, site activation, and data cleaning are different, because they often set the time to readout. That is where Formation Bio has aimed its operational tools: Apollo for study startup and enrollment monitoring, and Muse, developed with Sanofi and OpenAI, which speeds up recruitment strategy and materials.4
Time is also where the money is. Take an illustrative Phase 2 program (hypothetical figures): a conventional plan reaches readout in 24 months at $30 million, and an AI-assisted plan reaches it in 18 months at $26 million. The six months may matter more than the $4 million. Those months shorten the time capital is tied up, preserve time on market under patent protection, and bring forward a licensing decision, or a decision to stop.
Claims of AI-driven efficiency should therefore be read on three separate measures: time saved on a task, time saved to readout, and total cost per study. They are frequently reported as if they were one.
Better evidence from the same asset
There is a third development advantage between choosing an asset and administering its trial: designing a study that resolves the right uncertainty. Better patient selection, dose strategy, endpoint measurement, or recruitment feasibility can improve the value of the readout without changing the molecule. AI can help here by mining prior trials and real-world data for the populations and endpoints most likely to give a clear answer. Whether the answer is the right one still depends on the clinical question being asked.
Selection matters more than administration
The larger prize is upstream. Formation Bio targets programs that may have stalled for financial or strategic reasons rather than fundamental scientific limitations: a sponsor ran short of cash, reprioritized, or could not justify the next study.5 Its Phyla platform supports evidence-grounded diligence, and its Delphi model predicts trial outcomes from clinical and biological data.6,7
The economics favor this layer. Depending on acquisition costs and the value of a success, a developer that raises the share of its programs reaching proof of concept from 30% to 36% (illustrative figures) may have a more consequential advantage than one that cuts trial administration costs by 20%. The first improves where development capital goes; the second improves how efficiently it is spent.
This is also where the claim is hardest to support. Large pharma, mid-sized biotechs, specialty pharma, and asset-focused investors have hunted for stranded assets for decades, and many now use AI in the same workflows. Screening more assets, or screening them faster, is not the same as choosing better ones.
Across more than 350 diligence assignments since 2009, we have seen the same problems change a recommendation again and again.8 Management presentations read more strongly than the clinical study reports behind them. Summaries of agency meetings are more optimistic than the agency's own correspondence. CMC is often the last domain to get attention and the first to cause trouble after a deal closes. And some assets clear the scientific and regulatory bar but fail on commercial terms. Broader search finds more candidates; it does not, by itself, find these.
What is actually new
The asset-centric model is not. Housing each program in its own entity and sharing a central development organization is a structure Roivant Sciences and others have used for years.3 The logic is sound: no full operating company per asset, flexible capital allocation, and exits structured program by program. The asset, not the company, is what gets valued.9
Within that model, the AI question is productivity. Can one organization run more programs without proportional headcount, catch problems earlier, and make decisions more consistently? These are plausible gains, and they are measurable. They also have a cost: Formation Bio carries engineering, data science, data licensing, and compute alongside its drug developers. A proprietary platform can pay off at portfolio scale. For many sponsors, commercial tools built into existing workflows will be the better trade.
The financing vehicle is the newer element. Asset-level financing is not new, but a drug developer raising a dedicated fund tied to its own development infrastructure is less common.1 It fits the strategy: if programs are underwritten at industry-average odds, as Formation Bio's CEO has argued they should be,10 the advantage lies in how many programs you can fund at sensible prices and how quickly they resolve, and a dedicated fund, combining outside capital with Formation Bio's own commitment, lets it take more of them than its balance sheet alone would support.
Process evidence is not outcome evidence
Formation Bio does have a track record. It took BLKR201, a TYK2 inhibitor licensed in December 2025 with FDA IND clearance already in hand, from deal close to first participant dosed in five months.11,12 That is evidence of execution from a ready starting point rather than a compressed path to the clinic. Speed can also reflect indication choice, geography, or trial design rather than the tools. Separately, Sanofi licensed gusacitinib, which Formation Bio had taken into Phase 3 for chronic hand eczema, to study it in a new indication, in a deal worth up to €545 million in upfront and milestone payments plus royalties.13 That is evidence of partnering traction.
They are not yet evidence that AI selects better assets or raises success rates, and Formation Bio does not claim otherwise. Its CEO treats any predictive advantage as upside rather than a condition of viability.10 That makes the near-term case one of execution and capital efficiency: more well-priced shots at average odds, resolved faster. Demonstrating a predictive edge would take a portfolio, years of readouts, and comparison against similar programs run conventionally. Formation Bio's Delphi work moves in that direction by recording time-stamped predictions before readouts.7 Company-reported examples are a start; calibrated performance across a defined sample, and evidence that the model improves investment decisions, would be the stronger test.
An assessment of any AI-enabled developer should establish the following.
| Area | Measures | What it should establish |
|---|---|---|
| Asset selection | Screening volume, diligence time, decision outcomes against comparable deals | Better choices, not just more of them |
| Trial design | Protocol amendments, enrollment feasibility, endpoint acceptance | Studies that are more executable and more informative |
| Trial execution | Startup time, enrollment rate, time to database lock | Gains on the critical path, not only on tasks |
| Organization | Programs per employee, shared costs, external spend | Economies of scale net of technology cost |
| Outcomes | Milestones met, time and capital to proof of concept | More clinical evidence per dollar |
| Durability | Proprietary data, validated workflows, switching costs | An edge that holds as competitors adopt AI |
No single row is sufficient. Productivity per employee can rise while total cost rises with it, through CRO, specialist, or infrastructure spend.
Where AI helps diligence, and where it can mislead
Several of the diligence problems described earlier are, at root, cross-checking problems. Does the deck match the clinical study report? Does the summary match the FDA minutes? Does the license chain hold? These checks are exhaustive and tedious, and under deal timelines they are the first to be compressed.
AI can make systematic cross-checking far less laborious, so a more complete review fits within a normal deal timeline. That raises the floor of diligence for many buyers, not only AI-native ones. Doing this on a data room depends on contractual permission to process the material, an approved environment, and controls on access, retention, and use. Data governance is becoming part of diligence quality. AI-generated weighting, including model scores, still needs scrutiny and accountable judgment: which discrepancy changes the value, and whether a technically sound asset can earn a return at the price asked.
A prediction model produces a number whether or not the evidence supports one. A probability of success is useful when it rests on relevant, calibrated precedent. For a novel mechanism or modality with few comparators, it can look rigorous while describing very little.14
The more valuable contribution is narrower: better information at decision points. Analysis showing that an efficacy signal is confined to a small subgroup, that an endpoint fits the indication poorly, or that a plan rests on an untested regulatory assumption can change a valuation, a trial design, or whether to invest at all. Deciding which uncertainties can be modeled, which need new data, and which cannot yet be quantified remains the work of experienced clinical and scientific judgment.
The question that matters
The test for Formation Bio, and for any AI-native developer, comes down to one question.
Give an experienced pharma BD team, a competent CRO, and an asset-focused fund comparable AI tools: what can you do better, faster, or cheaper than they can, and where is the quantitative evidence?
Until that has a clear answer, the industry should keep three outcomes apart: doing existing work faster, making better development decisions, and achieving better clinical and financial results. Each has value. They are not interchangeable, and the first is far easier to demonstrate than the third.
What this means for asset holders, investors, and BD teams
AI is becoming standard in diligence and development. For many readers, the practical question is not whether to adopt it but how to work with counterparties who already have.
Biotechs holding stalled or deprioritized assets
Buyers like Formation Bio are screening more assets, faster, with AI-assisted evidence review. That could bring greater attention to programs that stalled for strategic or financial reasons, while raising expectations for the completeness and consistency of the supporting data package, because gaps between the pitch and the source documents are now cheaper to find. Reviewing the program the way an incoming diligence team would, before going to market, surfaces issues while there is still time to address them.
Investors, including new entrants to asset-level investing
Backing a portfolio of asset companies through a pooled fund asks for two kinds of assessment: the development portfolio itself, and the contribution claimed for the AI platform. Specialist capacity in clinical, regulatory, and CMC evaluation is what makes the second question answerable. The measures in the table above are a reasonable starting framework.
Pharma and biotech BD teams
AI widens the search across public literature, trial registries, and deal records, and teams that use it will see more of the opportunity set. The edge moves toward what a broader search cannot supply on its own: sourcing hypotheses, access and relationships, and the judgment to tell a real signal from a well-presented narrative.
The organizations that gain the most from these tools will be the ones that know which questions the tools cannot yet answer on their own.
Frequently asked questions
AI shortens a trial only when it speeds up the activity that controls the schedule, such as patient recruitment, site activation, or data cleaning, or protocol and regulatory drafting when that work is holding up study start. If enrollment is the constraint, faster document preparation saves labor but does not move the readout date.
A $600 million fund, announced October 28, 2026, to acquire or license 10 to 15 drug programs over three to five years. Each program sits in its own company and draws on Formation Bio's shared development and AI infrastructure. Formation Bio is the fund's general partner through a subsidiary and invests its own capital alongside limited partners that include research foundations, endowments, hospitals, and philanthropic organizations.
Process evidence shows that a team or system performs a defined task well, such as moving a licensed asset to first patient dosed quickly. Outcome evidence shows that using AI improves asset selection, clinical success rates, or returns. Outcome evidence takes a portfolio, years of readouts, and comparison against similar programs run conventionally.
AI can widen and speed up screening and surface gaps in the evidence, but screening more assets is not the same as choosing better ones. Formation Bio's CEO has argued for underwriting programs at industry-average probabilities of success and treating any predictive advantage as upside. Better selection may be worth more than lower trial administration costs, but it is the hardest benefit to demonstrate.
Keep three outcomes apart: doing existing work faster, making better development decisions, and achieving better clinical and financial results. Then test each area: asset selection against comparable deals, trial design and execution on the critical path, organizational productivity net of technology cost, time and capital to proof of concept, and whether the edge holds as competitors adopt AI.
As AI tools become widely available, access to them stops setting companies apart. The advantage moves toward proprietary data and validated workflows, sourcing hypotheses and relationships, judgment about the quality of evidence, and the capital structure that determines how many programs a developer can fund.
References
- Formation Bio. Formation Bio announces $600 million debut fund to expand drug development capacity. Formation Bio. Published October 28, 2026. Accessed November 3, 2026. https://www.formation.bio/press/formation-bio-announces-usd600-million-debut-fund-to-expand-drug-development-capacity
- Liu B. Introducing the $600M Formation Bio Drug Development Accelerator Fund. Formation Bio. Published October 28, 2026. Accessed November 3, 2026. https://www.formation.bio/blog/introducing-the-usd600m-formation-bio-drug-development-accelerator-fund
- Feldman A. This Sam Altman-backed $1.8 billion startup bets AI can get drugs through clinical trials faster. Forbes. Published April 16, 2026. Accessed November 3, 2026. https://www.forbes.com/sites/amyfeldman/2026/04/16/this-sam-altman-backed-18-billion-startup-bets-ai-can-get-drugs-through-clinical-trials-faster-formation-bio/
- Formation Bio. Introducing Muse. Formation Bio. Published November 12, 2024. Accessed November 3, 2026. https://formation.bio/blog/introducing-muse
- Feldman A. Why an AI biotech startup launched a $600 million fund to buy stalled drugs. Forbes. Published October 28, 2026. Accessed November 3, 2026. https://www.forbes.com/sites/innovationrx/2026/10/28/why-an-ai-biotech-startup-launched-a-600-million-fund-to-buy-stalled-drugs/
- Formation Bio. Phyla: grounding drug diligence in cited, contextual evidence. Formation Bio. Published October 2026. Accessed November 3, 2026. https://www.formation.bio/blog/phyla
- Formation Bio. Introducing Delphi: Formation Bio's clinical trial prediction model. Formation Bio. Published June 3, 2026. Accessed November 3, 2026. https://www.formation.bio/blog/delphi-clinical-trial-prediction-model
- Walker A. Due diligence: the pinnacle of expertise-based consulting. Alacrita. Accessed November 3, 2026. https://www.alacrita.com/whitepapers/due-diligence-the-pinnacle-of-expertise-based-consulting
- Rodriguez L. The biotech company is not the product. Alacrita. Published June 2026. Accessed November 3, 2026. https://www.alacrita.com/blog/early-stage-biotech-evidence-before-infrastructure
- Liu B. Uneven frontiers. a16z. Published June 17, 2026. Accessed November 3, 2026. https://www.a16z.news/p/uneven-frontiers
- Formation Bio. Formation Bio doses first participant in phase 1 trial of BLKR201, a CNS-penetrant TYK2 inhibitor. Formation Bio. Published June 9, 2026. Accessed November 3, 2026. https://www.formation.bio/press/formation-bio-doses-first-participant-in-phase-1-trial-of-blkr201-a-cns-penetrant-tyk2-inhibitor
- Formation Bio. Formation Bio licenses phase 1-ready CNS penetrant TYK2 inhibitor from Lynk Pharmaceuticals. Formation Bio. Published December 10, 2025. Accessed November 3, 2026. https://www.formation.bio/blog/formation-bio-licenses-phase-1-ready-cns-penetrant-tyk2-inhibitor-from-lynk-pharmaceuticals
- Sanofi signs €545M deal to explore Formation's JAK/SYK inhibitor in new indication. Fierce Biotech. Published June 2025. Accessed November 3, 2026. https://www.fiercebiotech.com/biotech/sanofi-signs-545m-eur-deal-explore-formations-jaksyk-inhibitor-new-indication
- Walker A, Hepler C. When probability is fiction: risk vs. uncertainty in biotech due diligence. Alacrita. Published January 9, 2026. Accessed November 3, 2026. https://www.alacrita.com/blog/risk-vs-uncertainty-biotech-due-diligence
Contact Alacrita
Send a short note and we will be in touch.