AI agents

We’ve watched generative AI learn to design proteins and RNA. The oldest, hardest target is the small molecule — chemical space is astronomical and the bench is unforgiving. This year brought both: a target-aware 3D diffusion model whose drugs actually worked in a wet-lab assay, and an unusually honest benchmark that measures the gap between a leaderboard win and a real hit. Read together, they tell you exactly where the field is.

Self-driving labs used to run the experiment a human designed. The 2026 shift is subtler and bigger: the agent now decides which experiment to run next — and the bottleneck moves to physical validation. That reframes the frontier as compressing wet-lab rounds and building a safe, standard way for reasoning agents to drive real instruments — exactly the layer our lab builds.

Today in AI for life science: a model reads a cheap histology slide and predicts where thousands of genes are expressed, a foundation model puts dissociated cells back in their tissue, and an autonomous agent runs the whole spatial-analysis workflow — with the usual caveat about who checks the answer.

Today in AI for life science: an open, model-agnostic science workbench arrives as a rival to proprietary tools, a real open-vs-closed contest forms over the scientific AI stack, and why the lab’s whole bet is on open.

Today in AI for life science: models move past static structures to predict proteins as moving ensembles, that frontier hits a data wall, and self-evolving agents repurpose drugs for rare diseases — where the wet lab both confirms and refutes.

Today in AI for life science: vision-language models turn microscopes into conversational instruments, a reality check on the gap between seeing and understanding an image, and multimodal pathology models fusing slides with clinical text at scale.

Today in AI for life science: a new protocol lets AI agents safely drive lab instruments, autonomous labs run experiments at a scale that outpaces the rules, and foundation models reach a new frontier — the microbiome.

Today in AI for life science: a general-purpose ‘AI biologist’ matches human researchers across 25 subfields, cell models generalize across species, and the newest frontier model ships bio-tuned — and on a government leash.

Today in AI for life science: a cell-language model proposes a cancer lead that holds up in the wet lab, agents turn plain-language requests into reproducible image analysis, and microscopes start deciding what to image themselves.

Today in AI for life science: the big models get plumbed into real scientific data via connectors, AI moves into RNA drug design, and machine learning starts designing the CRISPR edit itself.