Robots, agents, and the lab bench

Agentic Discovery  ·  Nº Eleven  ·  08 May 2026

Editor's Note

Today's run leans heavy on the wet-lab side — robots running proteomics, compilers checking protocols, and treatment planners that grade themselves.

Nº 01 · The Lede  —  bioRxiv  —  Agents · Infrastructure

Robotic proteomics meets AI agents

Fig. I  bioRxiv, 08 May 2026.

Robotic perturbation proteomics pairs automated sample handling with AI agents to scale drug mechanism-of-action discovery, per a new bioRxiv preprint. The setup runs perturbation experiments on a robot, then hands proteomic readouts to agents that propose targets, design follow-ups, and iterate. Early results show the loop recovers known MoAs and surfaces candidates traditional screens miss. Marks one of the first credible demonstrations of closed-loop AI-plus-wetlab discovery at meaningful throughput.

Read the source →

Nº 02  —  arXiv  —  Agents · Infrastructure

Fig. II  arXiv, 08 May 2026.

TheraAgent grades its own treatment plans

TheraAgent self-improves on therapeutic planning by critiquing its own outputs against guidelines and patient context, then rewriting until the plan clears its internal rubric. The arXiv paper reports gains over single-pass LLM baselines on precision and comprehensiveness. Pushes treatment-planning agents toward a quality bar that doesn't depend on a human in the loop for every draft.

Read more →

Nº 03  —  bioRxiv  —  Field report

Fig. III  bioRxiv, 08 May 2026.

Compiler-verified lab protocols

Compiler-verified protocols give AI-generated lab procedures a static-analysis layer before they touch a robot, catching impossible volumes, missing reagents, and unsafe steps at compile time rather than mid-run. The bioRxiv work proposes this as the missing safety substrate for autonomous biology. Raises the floor for what an agent-driven wet lab should ship — protocol verification becomes table stakes, not optional.

Read more →

Also Filed  ·  Two Briefs from the queue

Nº 04  —  arXiv  —  Agents · Infrastructure

Multi-agent system for translational medicine

BioResearcher coordinates specialist agents across literature review, hypothesis generation, and experiment planning for translational medicine, guided by clinical-scenario prompts. Adjacent to story 1's loop but oriented at the bench-to-bedside seam rather than MoA discovery — narrows the gap between LLM-driven research planning and clinical relevance.

Read →

Nº 05  —  Anthropic  —  Field report

MCP, the standard nobody can skip

Anthropic's Model Context Protocol (MCP — an open spec for plugging AI assistants into data systems and tools) keeps spreading as the default agent-to-tool interface. For biology, that means agents inherit a shared plumbing layer for ELNs, instrument APIs, and databases without bespoke glue per vendor.

Read →

· · ·

Reply with your discoveries. A human reads them. Forward freely.