By Hamza Faisal

Google DeepMind has taught its Antigravity platform to do chemistry. The company announced new Cheminformatics Skills for the agentic AI workbench on November 3, turning it into a research assistant that can search thousands of compounds, analyse their structures and propose entirely new drug-like molecules — in about an hour.

In a demonstration built around real lung cancer treatments, an Antigravity agent worked through data on crizotinib and related compounds, ran matched molecular pair analysis and suggested novel rigidified macrocycles, complete with proposed synthesis routes. A typical iteration of that workflow takes multiple days of manual work. The agent finished it in roughly sixty minutes.

From days of data wrangling to an hour

The point, DeepMind says, is to let scientists focus on hypotheses rather than data wrangling. The agent pulls molecules out of research documents, links each one to the data reported alongside it and prepares the compounds for downstream analysis, all through natural-language prompts. The language model coordinates the chemistry steps while standard software handles the molecular calculations.

Macrocycles — large ring-shaped molecules — are a promising but difficult class of drug candidates. Their rigid, ringed structure can help them bind tightly to disease targets, but designing them is slow, fiddly work. In the demo, the agent explored how rigidifying a floppy molecule changes its potency, then proposed new ring structures a medicinal chemist could plausibly make.

Real chemistry tools under the hood

The skills are grounded in RDKit, the open-source cheminformatics toolkit built over two decades by a global community of contributors, alongside connectors to curated chemical resources including ChEMBL and PubChem. That grounding matters: the agent is not free-associating molecules from training data, but working from established chemistry software and real compound databases.

Google DeepMind describes Antigravity with the new skills as a collaborative science workbench, built for both computational chemists and experimental bench chemists. It is the platform's second documented scientific workflow, following earlier work using Antigravity to search for antimicrobial peptides.

Free access for 10,000 scientists — with a catch

The Cheminformatics Skills are available now. Individual users can download the Antigravity app and install the Science plugin, while enterprise users can start with a business account and Google Cloud credentials. Google is also offering up to 10,000 eligible scientists its highest usage tier free for six months, with rolling onboarding subject to eligibility and capacity.

There is an important caveat. Google stresses that the agent's designs are testable hypotheses, not finished drugs. Sparse data, noise between laboratories and narrow measurement ranges all limit how precise the predictions can be, and every proposed molecule must still be synthesised and tested in a laboratory before anyone knows whether it works.

Even so, compressing a multi-day compound search into an hour changes the economics of early drug discovery. If the lab results hold up, the slowest part of finding a new medicine may soon be the chemistry itself, not the search for where to start.