From predictive models to agentic AI drug discovery
AI drug discovery now refers to connected, autonomous agents that translate documents, design experiments and trigger lab workflows, creating continuous loops between digital models, physical experiments and clinical data rather than isolated prediction tools. These AI agents automate repeatable steps such as protocol standardization, retrosynthesis planning and virtual screening, and then call laboratory systems or domain experts when needed. The shift marks a move from static machine-learning models to agentic workflows that can coordinate tasks across clinical trial protocols, autonomous lab chemistry and drug screening agents. Instead of focusing only on predicting a molecule’s activity, research teams are wiring AI directly into how they plan studies, run assays and manage data handoffs. This architecture makes it possible to scale drug discovery processes while keeping scientists in charge of scientific judgment and regulatory decisions.
Clinical trial protocols go machine-readable
In clinical development, protocol documents have long been a bottleneck because every change ripples through eCOA apps, data standards and trial documentation. Medable’s Digital Data Flow Agent addresses this by turning static clinical trial protocols into structured data in CDISC USDM 4.0 JSON, so downstream systems can update automatically instead of relying on manual re-keying. Earlier, the company used AI to translate a study protocol directly into a configured eCOA mobile app with questionnaires, workflows and about 25 language translations, reporting that this could halve deployment timelines that used to run 12 to 16 weeks. The new agent extends that idea to broader protocol-to-machine-readable conversion, a key step toward more real-time trials. By letting AI handle these clinical trial protocols, Medable says trial experts can oversee more sites and studies while focusing on high-value decisions rather than repetitive document work.
Retrosynthesis and autonomous lab chemistry connect design to the bench
On the chemistry side, autonomous lab chemistry is starting to connect AI design directly to wet-lab execution and biomanufacturing. At MilliporeSigma, AIDDISON and related AI capabilities are being wired into a broad portfolio that spans research chemicals, reagents, filtration, water systems, software and automation. The company’s Opentrons-powered AAW Automated Assay Workstation already runs routine assays with less manual handling, while greener, bio-based solvents provide drop-in replacements for long-used HPLC chemicals such as acetonitrile. This mix of AI retrosynthesis, automation and sustainable reagents underpins an autonomous lab stack that can close the loop between digital route design and physical synthesis. Karen Madden describes AI and generative AI as “unlocking events” that open up practical possibilities in areas like novel modalities, including ADCs, mRNA therapies and PROTAC therapies, turning what used to be isolated tools into coordinated drug discovery pipelines.
Drug-discovery APIs built for agent interaction
Beyond labs and clinics, a new layer of drug-discovery APIs is emerging that is explicitly built for coding agents as well as human users. Boltz Bio’s platform combines BoltzProt-1 for protein design with BoltzMol-1 for small-molecule hit discovery and exposes them through an API, SDKs and integrations with coding assistants such as Claude Code, Codex and Gemini CLI. At launch, the company noted that its own scientists had already been accessing these models through coding agents, so it “built this for agents as much as for people.” Experimental validation across 10 targets, including GPCRs, kinases, ion channels and protein–protein interactions, shows confirmed binders at the predicted sites, positioning these tools as reliable drug screening agents. With commercial biomolecular models unevenly available, such APIs let teams plug agentic workflows directly into screening and compound analysis without building entire GPU pipelines in-house.
Toward fully agentic discovery workflows
Taken together, these advances point toward discovery pipelines where AI agents orchestrate work across protocol standardization, lab execution and digital screening. In clinical R&D, protocol-to-JSON translation agents compress setup times and support continuous data review. In chemistry, AI-driven retrosynthesis and autonomous assay workstations turn proposed routes and screens into physical experiments, while greener reagents keep production practical and sustainable. On the digital side, drug screening agents interact with biomolecular APIs, automatically ranking hits and requesting follow-up predictions. The industry is moving beyond standalone predictive analytics toward fully agentic workflows, where AI coordinates many small tasks and humans focus on scientific interpretation, risk assessment and strategy. That transition will not remove human oversight, but it is already redefining how many molecules and trials a single R&D team can handle at once.
