---
title: "Google adds chemistry skills to Antigravity, with lab validation still essential"
description: "Google adds Cheminformatics Skills to Antigravity. Explore Science plugin access, the research demonstration and why experimental validation still matters."
url: "https://promptszone.com/google-antigravity-cheminformatics-skills-science-plugin"
published: "2026-11-04T01:00:07Z"
categories: ["AI News"]
---

# Google adds chemistry skills to Antigravity, with lab validation still essential

New chemistry tools connect AI workflows with structured analysis, while proposed compounds still need experimental checks.

![Official Antigravity screenshot showing an assistant alongside compound-analysis charts from Google’s chemistry demonstration.](https://promptszone.com/assets/uploads/2e660ced22d268b4e6b2d227ce8c1ba3.jpg)

**Summary:**

Google’s Science plugin adds cheminformatics workflows. Availability, evidence limits and practical questions for evaluating the demonstration.

Google is putting chemistry-specific tools inside Antigravity. Its [November 3 announcement](https://deepmind.google/blog/agentic-cheminformatics-with-google-antigravity/) describes Cheminformatics Skills that coordinate literature searches, compound analysis and proposed molecular designs. This is a workflow update, not a new foundation-model launch.

## What is available

Google says the skills are available through the Antigravity app’s Science plugin. The system combines language-model orchestration with deterministic chemistry tools, including RDKit, and connections to PubChem and ChEMBL. Its demonstration moves from literature-derived compounds to predictive analysis and candidate designs in about an hour, according to Google.

The [Science page](https://antigravity.google/use-cases/science) links the plugin’s open-source customizations and an application for free access. The announced offer covers 10,000 scientists for six months at the highest usage tier, subject to eligibility, capacity and selection. It is not an automatic free upgrade for everyone.

## A faster hypothesis is not a validated medicine

Google explicitly warns that noisy assays and sparse data limit predictions. Suggested compounds still require synthesis and experimental validation. The demonstration does not establish clinical effectiveness or provide a controlled comparison against competing AI systems.

## How to evaluate the workflow

**Our assessment:** the useful question is whether a team can audit the work, not simply whether the agent produces an impressive dashboard. A sensible pilot would start with a familiar dataset and a result the researchers can independently check.

- **Trace each result.** Keep the original source, extraction decisions and transformation history alongside each output. A polished chart should not hide where its measurements came from.
- **Separate prediction from observation.** Mark model estimates distinctly from experimental measurements, and keep units and assay conditions visible.
- **Test reproducibility.** Save the code, package versions and data snapshot. Ask another researcher to rerun the analysis without relying on the original conversation.
- **Measure review effort.** Include time spent correcting data and checking outputs when comparing the agent-assisted workflow with the existing process.

These are suggested evaluation criteria, not results from a PromptsZone test. We have not independently reproduced Google’s demonstration.

The same distinction between a promising demonstration and verified outcomes also matters in [AI-assisted physical engineering](https://promptszone.com/openai-physical-engineering-research-manufacturing-checks).

Featured image: unaltered Google Antigravity screenshot from Google DeepMind’s November 3 demonstration. Source: Google. It illustrates the software, not independently validated findings. Google and Google Antigravity are trademarks of Google LLC. This article is not affiliated with or endorsed by Google.