Skip to content

AI

Google's lab agent drafts new drug candidates in about an hour

Google DeepMind taught its Antigravity agent some chemistry. In its lung cancer demo, the loop that normally eats days runs in roughly an hour, and the carbon-based chemists get to supervise.

Cover photo: Dietmar Rabich / CC BY-SA 4.0

TL;DR

  1. DeepMind adds Cheminformatics Skills to its Antigravity agent for drug discovery workflows.
  2. Demo on real lung cancer treatments: about one hour, versus multiple days per iteration.
  3. No word on EU availability, so European labs should not clear their calendars yet.

What happened

Google DeepMind published a new set of Cheminformatics Skills for Google Antigravity, its AI agent. The post is dated November 3, 2026, and is credited to Zoi Tsangalidou, Ruijie (Kelly) Zhang, Sukhdeep Singh and Michael V. LeVine.

The pitch: designing a new medicine means pulling messy data out of charts, figures and literature, training predictive models, and then proposing molecules worth making. DeepMind says one pass through that loop typically takes multiple days.

With the new skills, the agent handles the loop itself. In the demo, built on real-world lung cancer treatments, it searches thousands of compounds and runs matched molecular pair analysis. It then proposes novel rigidified macrocycles, complete with synthetic routes.

That is DeepMind's own demo and DeepMind's own number. Nobody outside the building has published a stopwatch yet.

Why it matters

Chemistry has a data-wrangling problem. DeepMind itself says chemists spend too much time wrestling data, partly because one molecule can be written down in many different ways. That is the boring, expensive part, and it is exactly what agents are good at.

The interesting shift is scope. This is not a model answering a chemistry question. It is an agent chaining the whole job: find the data, analyze it, build the models, propose candidates, and sketch how to synthesize them.

Days per iteration down to an hour, if it holds beyond a showcase, means more rounds of trial and error in the same week. In drug discovery, more iterations is the whole game.

Caveat, because we are not the vendor: a demo on curated lung cancer examples is not a pipeline in production. Proposing a molecule on screen and making it work in a lab are very different things.

What to do with it

  • Read the source post.DeepMind's write-up shows the workflow end to end, which tells you what the agent actually does versus what the headline implies.
  • Map your own loop.If you work in chemistry or biotech, list which steps eat days: data curation, model training, candidate proposals. Those are the targets.
  • Keep the human checkpoint.Treat proposed molecules and synthetic routes as drafts for expert review, not orders.
  • Watch for availability.Nothing in the announcement settles access or regional rollout, so check before planning around it.

Ciao, sapiens. The agent can draft the molecule, but it still cannot hold the pipette. Enjoy that while it lasts.