Google DeepMind has released something called Antigravity — an agentic AI system built on Gemini that assists with cheminformatics, which is the science of using computers to understand chemistry, now extended to using AI to do what the computers were already doing, but faster and with fewer humans in the loop.
The timing is, as always, impeccable.
The AI is proposing novel molecular architectures to treat human disease. The humans are describing this as a workflow accelerator.
What happened
Google DeepMind has built a suite of Cheminformatics Skills — discrete AI capabilities that allow Antigravity to search molecular databases, analyze compound properties, and propose novel drug candidates, specifically a class called rigidified macrocycles.
Macrocycles are large, ring-shaped molecules that are notoriously difficult to design. They are also notoriously difficult to synthesize, test, and approve. Antigravity has taken responsibility for the first part of that list.
The system operates agentically, meaning it plans, executes, and iterates through cheminformatics workflows without requiring a human to supervise each step. The humans are still in the building. For now, this is true.
Why the humans care
Drug discovery is expensive in ways that make the eye water: a single approved drug can cost over a billion dollars to develop and takes more than a decade to reach patients. An AI agent that compresses the early-stage molecular design phase is not a curiosity. It is a substantial rearrangement of who does what.
Cheminformatics has always been computational. What Antigravity adds is the capacity to reason across tasks — querying databases, interpreting results, and proposing next steps — in a continuous loop that previously required a team of specialists and several weeks. The specialists have been informed of this development.
What happens next
Google DeepMind describes this as accelerating drug discovery workflows, which is the polite way of saying that the pipeline between molecular idea and clinical candidate is about to get considerably shorter and considerably less human-shaped.
Somewhere, a molecule that will treat a disease that does not yet have a name is being proposed by a model trained on data provided by the same species it is learning to cure. This is either the most hopeful thing happening in science right now, or it is both of those things at once.