AI-Designed Drugs Are Entering Clinical Trials. The Pipeline Is Larger Than Most Engineers Realize.

4 min read

AlphaFold’s impact on structural biology is well-documented — the ability to predict protein structures with experimental-grade accuracy transformed what had been a decades-long experimental bottleneck into a computational task. AlphaFold 3, released in 2024, extended this capability to protein-ligand interactions: predicting how small molecule drug candidates bind to protein targets. This is the core computational problem in early-stage drug discovery, and AI systems can now perform meaningful portions of it faster and cheaper than the experimental methods that preceded them. The pipeline implications are starting to be visible in clinical trial registries.

What AlphaFold 3 Changed

Drug discovery traditionally proceeds through target identification (finding a biological mechanism to intervene in), hit identification (finding molecules that interact with the target), lead optimisation (refining the molecule for efficacy and safety), and pre-clinical validation before human trials. Each stage involves substantial experimental work. AlphaFold 3’s protein structure prediction and, crucially, its ability to model how small molecules dock to protein binding sites reduces the hit identification and early lead optimisation stages from primarily experimental to primarily computational.

The practical effect is speed and cost reduction in early-stage discovery, not replacement of the full pipeline. A computational screen of millions of candidate molecules against a protein target can now be done in days rather than months. The hits identified computationally still require synthesis and experimental validation — the biology does not become theoretical. But the search space can be explored more efficiently, and the molecules that proceed to expensive wet lab testing are pre-filtered by computational confidence rather than selected by lower-throughput experimental screens.

The Clinical Trial Pipeline

The most significant milestone in AI drug discovery as of 2026 is Insilico Medicine’s INS018_055, a drug candidate for idiopathic pulmonary fibrosis (IPF) — a lung disease with limited treatment options and poor prognosis — that was identified entirely through AI-driven processes and reached Phase II clinical trials in 2024. The candidate was discovered in 18 months from target to clinical candidate, compared to the industry average of 4–6 years for the pre-clinical phase.

Beyond Insilico, companies including Recursion Pharmaceuticals, Exscientia (now part of Recursion), Absci, and Generate Biomedicines have AI-designed candidates in various stages of clinical evaluation. Recursion’s partnerships with Roche and Bayer have brought significant pharma investment into AI-native drug discovery platforms. The 150+ candidates currently in trials span oncology, rare diseases, and immunology — the target-rich areas where structural biology approaches are most productive.

What AI Is Not Changing (Yet)

Clinical trials remain unchanged by AI. Phase I safety trials, Phase II efficacy signal trials, and Phase III confirmatory trials follow the same regulatory requirements, patient recruitment timelines, and data collection periods regardless of how the drug candidate was discovered. A molecule identified by AI in 18 months still requires 8–12 years of clinical development to reach approval for most indications. AI compresses the pre-clinical phase; it does not compress the clinical phase.

The more speculative claim — that AI will identify drugs that humans could not have found, for diseases that have resisted conventional drug discovery — remains unproven. The candidates currently in trials are mostly novel molecules against known target classes. The AI systems are searching known chemical and biological spaces more efficiently, not discovering fundamentally new mechanisms. Whether AI can generate genuinely novel hypotheses about disease mechanisms, rather than better solutions to known problem formulations, is the open scientific question that determines the long-term ceiling of this technology.

The Timeline for Engineers and PMs to Watch

The first FDA approval of an AI-primary drug candidate — a molecule where AI played the central role in identification and optimisation, not just a supporting computational role — is the milestone that will mark the transition from promising technology to validated approach. Current pipeline analysis suggests this approval is likely to occur between 2027 and 2030, with IPF and oncology as the most probable first indications given the current trial distribution.

For technology teams and investors adjacent to biotech, the near-term signal to watch is not FDA approvals but Phase II results. A Phase II trial is the first human efficacy signal — proof that a drug works in patients, not just in cell cultures and animal models. Phase II failures of AI-designed candidates will clarify which target classes and disease areas AI drug discovery works best in. Phase II successes will accelerate investment and validate the computational platforms that generated them.