Google Antigravity — searchable YouTube transcripts & AI Q&A

Accelerating antimicrobial peptide discovery with Google Antigravity

Google Antigravity can accelerate antimicrobial peptide discovery by screening proteomes, validating novelty, and predicting structure and function in one workbench. The agent identified a novel lactococcin 972 family bacteriocin from Streptococcus pseudopneumoniae, confirmed its antimicrobial fold via AlphaFold and Foldseek, and generated a wet-lab experimental plan. This approach demonstrates how agentic workflows can streamline early-stage drug discovery, though the candidate still requires experimental validation.

Building Dashboards: Google Antigravity Livestream

The speaker demonstrates how to build live, shareable dashboards with Google Antigravity, using data from Salesforce, BigQuery, and Jira, and deploying them as web apps via Google Apps Script. He argues that this approach is superior to sending static HTML files because it provides live data, identity-aware access, and personalized queries. He also shows how to use automations and skills to keep the dashboard updated and even fix customer issues automatically.

How to Build an Android App on Antigravity

The speaker demonstrates how to build an Android app entirely within Antigravity, using Stitch for screen design and the Build with Google Android plugin to run it on an emulator or a real device. The key workflow is to export screens from Stitch, set up the Stitch MCP server in Antigravity, and then prompt the agent to build the app with native Jetpack Compose views, which can be verified and tested interactively.

New Updates in Google Antigravity | Episode 8

Google Antigravity's latest update introduces Gemini 3.8 Flash as a daily workhorse, alongside specialized multi-agent commands: teamwork for long, frontier research tasks and boost for focused problem-solving. The update also enables interactive visualizations and UI prototypes, a permission system that auto-runs safe commands, and support for local AI models via Light RT for offline, private execution.

Planning and Artifacts: Google Antigravity Livestream

The speaker argues that separating planning from implementation is key to using AI agents effectively, allowing you to think and research on the go while agents handle the building. He demonstrates a workflow where he creates GitHub issues via a custom skill, refines them remotely, and then uses a work tree and boost mode to implement the plan, reviewing changes in the context of the issue. He also notes that medium thinking effort is a good default, and that flash models are fast enough for most tasks.