Antigravity, Claude Code and Copilot: choosing without marrying any of them

In a year we have gone from “the editor suggests the next line” to “the agent opens a branch, writes, runs the tests and hands you the result”. It is worth understanding how the options genuinely differ before standardising.

Three philosophies, not three brands

• Copilot was born inside the editor: you code and the tool helps. Its advantage is that it does not change how the team works, and in a Microsoft shop it fits without friction.
• Claude Code lives in the terminal and treats the work as a conversation with access to your repository: it reads, edits, runs and explains. It suits people who already live on the command line.
• Antigravity, from Google, inverts the framing: instead of putting the agent inside the editor, it puts the editor inside a system built for agents. It has a classic editor view and a manager view where multiple agents are launched and monitored in parallel, and it produces artifacts — plans, task lists, screenshots, browser recordings — so you can verify what happened without reading every tool call.

One detail that surprises people: all three support models from several vendors. Antigravity offers Gemini 3, Claude Sonnet 4.5 and GPT-OSS. Choosing a tool and choosing a model stopped being the same decision.

What actually matters

Feature comparisons age in weeks. These four questions do not:

• Where does your code go? Who processes it, where it is stored, and whether it is used for training. If you work with regulated clients this is not negotiable and comes before any trial.
• What can the agent touch without asking? Writing files, running commands, opening the browser, reaching the network. Autonomy is the feature and also the risk.
• How do you verify what it did? An agent that delivers a plan and a screenshot gets reviewed in two minutes. One that delivers an 800-line diff with no context does not get reviewed: it gets accepted blind, which is worse than not using it.
• What happens if it disappears? In June 2026 Google retired Gemini CLI and the Code Assist extensions, replacing them with Antigravity. Anyone who had built their workflow on top had to migrate. That is not a criticism: it is the pace of this market.

Standardise on practices, not tools. Practices — mandatory review, tests before merge, a branch per task — survive the next product change. The tool may not.

Where we stand

The same as with the rest of the stack: depth where it adds most value, without tying ourselves down. We use Claude day to day because it fits how we work, and we are not going to tell you it is the universal answer. If your team lives in Visual Studio and Azure DevOps, Copilot will give you more output for less friction, and that is worth more than any comparison.

What we will say: try two of them on a real, boring task — a small migration, a refactor with tests — not on a toy exercise. The difference between tools shows up in ugly code, not in the clean example.