Google's Antigravity coding agent used to be a thing you visited. Now it's a thing you can bolt directly onto your own product — no field trip required. This week Google shipped antigravity-preview-09-2026, and quietly turned a flagship agent into infrastructure.

From Standalone Tool to Callable API

The update brings the Antigravity agent's tools and behavior — the ones that let it plan, reason, run code, manage files, and search the web inside a secure Linux sandbox — straight into the Gemini API's Interactions API and AI Studio, running on Gemini 3.8 Flash. Alongside it, Google shipped two new supporting pieces: a Files API for moving data in and out of the agent's sandbox, and a Credentials API that lets the agent securely authenticate to services you already use, from MCP servers to third-party APIs.

The harness itself got upgrades too — better prompt caching, native code search tools, and more efficient line-range file editing, all available immediately via the Python or JavaScript SDK, or plain REST.

The Move That Matters More Than the Model

Model announcements are a dime a dozen these days, but shipping the agent's actual operating harness as an API is the more interesting move. It means any developer can now embed "an agent that plans, codes, and manages files in a sandbox" into their own product, instead of just using Google's version of it inside Google's own interface.

That's the real competitive story here: the fight has quietly shifted from "whose model scores higher on the benchmark" to "whose agent scaffolding is easiest to build a real product on top of." Credentials and Files APIs are the unglamorous plumbing that decides whether an agent stays a demo or becomes something a business actually ships.

The chatbot wars got all the headlines; the agent-infrastructure war is the one that'll actually reshape how software gets built.

If you're weighing how (or whether) to wire agentic tools like this into your own business's workflows, that's exactly the kind of project we'd love to help you think through.

Source: Google AI for Developers