What would you build if you started a Biotech today?

Direction is clear, timing uncertain

Most current writing on AI in healthcare falls into one of two camps: how pharma could use AI internally, or how patients, physicians, and payers are adopting AI on their own. For me, the more interesting question sits between them: What happens to pharma when adoption reaches scale?

AI use in obtaining medical advice is rising: 81% of physicians now report using AI in practice, up from 38% in 2023 (AMA, 2026). But scope varies widely, and whether a physician or patient turns to AI depends mainly on their perception of AI, less on technology capability per se. So advances in technology alone will not predict speed of adoption. No healthcare technology shift in recent history compares well to AI, and institutional friction (regulation, liability norms, physician access) may slow adoption but is unlikely to stop it. Timing is uncertain. Direction is clear.

Patients lead. Physicians follow. Payers accelerate.

The most likely driver is patients themselves. Waiting weeks for a specialist appointment while managing symptoms, adverse events, or diagnostic uncertainty is stressful. AI offers immediate, always-available guidance. The quality concern is real but temporary, and for non-specialist advice already fading. In primary-care therapeutic decisions, ChatGPT v3.5 was correct in 55.6% of cases vs. 54.3% for family physicians, incorrect in 5.2% vs. 11% (Simão et al., BMC Primary Care, 2025).

In real-world settings, AI is not yet ready. A recent study tested 21 frontier models (GPT-5, Claude 4.5) on clinical reasoning and found all failed to produce an appropriate differential diagnosis more than 80% of the time (Rao et al., JAMA Network Open, April 2026). Mount Sinai researchers separately found ChatGPT Health (OpenAI’s consumer health tool launched January 2026) undertriaged 51.6% of true emergencies, sending patients with diabetic ketoacidosis and impending respiratory failure to 24–48h evaluation rather than the ED (Ramaswamy et al., Nature Medicine, February 2026). Models excel when the data is complete; they struggle at the open-ended start of a case. But the gap between benchmark and bedside is narrowing, and for triage-level guidance most patients seek first, it may already be close enough.

Trust, separately, can lag use — patients rated identical advice as less reliable when labelled ‘AI,’ though the latest study on this is old in AI years given the pace of model improvement (Reis et al., Nature Medicine, 2024). And trust in AI builds with familiarity, much like trust in a new physician. A Yale physician, commenting on the Rao study mentioned above, reached a similar conviction: patients will demand AI reviews of physician plans (currently called “clinical decision support”, CDS), insurers will enforce it, and physicians reading today’s studies about LLM weaknesses may later wonder how they missed the signals (Wilson, Medscape, April 2026).

If patients adopt AI as a first step, physicians will follow. This is not a new dynamic. A study on direct-to-consumer advertising has shown that when patients request a specific treatment, physicians typically accommodate, often against their own judgment (e.g., Kravitz et al., JAMA, 2005). The effect may be less pronounced in specialty areas such as oncology, but AI provides a more sophisticated version of the same mechanism. 50% of physicians already report patients consult AI before their visit (Sermo, 2025). Payers accelerate this: current HCP systems are inefficient, and incorrect or late diagnoses are expensive. AI as a first-line triage tool and as a second opinion on costly prescriptions serves cost reduction, patient convenience, and may improve outcomes. Medicare’s WISeR model already uses AI to screen treatment requests across six US states; UnitedHealth is aiming for nearly $1 billion in AI-related savings this year (CMS, 2025; UnitedHealth, Q4 2025 earnings). Not all payers will move this fast, but again the direction is clear. The Healthy Technology Act, introduced in the US Congress in January 2025, even proposed to let AI qualify as a ‘practitioner eligible to prescribe drugs’ (unlikely to pass, but the fact it was introduced is itself a signal; Gilbert et al., npj Digital Medicine, March 2025).

Physicians who learn to collaborate with AI will gain relevance and influence, and preserve the human element of care. This may include, but isn’t limited to, examinations and procedures AI is (still) incapable of — a point Eric Topol has been making since Deep Medicine (Topol, Basic Books, 2019).

Evidence replaces share of voice

When AI influences or even decides which diagnostic tests are ordered and which drugs prescribed, established pharma industry communication channels will lose effectiveness.

Evidence and data quality become the primary competitive advantage. Prescribing influence shifts from physicians to algorithms. AI-powered CDS tools (UpToDate Expert AI, Epic’s embedded models, OpenEvidence at Mount Sinai) already analyze patient data against treatment guidelines before a prescription is written (e.g., Pharma Marketing Network, March 2026). The promise is a reduction in errors and hence improved patient outcomes. However, this hypothesis only holds if the AI systems themselves can distinguish relevant data from noise and promotion, prioritizing evidence quality over volume. Otherwise share of voice merely moves from physicians’ minds to training data. Over time, the models that win will be those that use evidence most rigorously, not those biased by high share of voice.

For pharma and biotechs, building a new “evidence communication engine” from scratch may be more viable than adjusting within existing structures. This approach gains appeal as the surrounding system also shifts — insights will be collected differently, market access and cost-effectiveness assessed differently by payers, and formularies built differently by providers. Across the system, AI use points toward more scientific rigor.

On the other hand, large pharma’s existing infrastructure like KOL and investigator networks, established sales forces, and omnichannel execution aren’t just expensive. They form a barrier to adaptation and may explain why pharma is being outpaced by the rest of the healthcare ecosystem. Menlo Ventures’ 2025 healthcare survey of 700+ executives captured the asymmetry: Health systems and outpatient providers are shortening AI procurement cycles, while pharma has stayed flat (Menlo Ventures, 2025). Biotechs with capital and no legacy can and will build for the coming world. What would you build if you started today?

About the author

Christian Wasmer

Co-Founder and Managing Partner, Consilience

Advises pharma and biotech on Medical Strategy, Clinical Planning, and New Product Planning, with therapeutic focus in oncology (heme and solid tumors), rare genetic and metabolic disorders, and cardiovascular disease. Over 10 years in strategic advisory, including Prescient Healthcare Group and Trinity Life Sciences, with earlier digital strategy work at the Boston Consulting Group. MSc and PhD (ETH Zurich) in physics and oncology/neurology research; Postdoc at Harvard Medical School.

Consilience is a boutique Life Science consultancy linking together science, strategy, and organizational context. We serve as embedded advisors to pharma and biotech leaders.

consiliencestrategy.com