Google Antigravity release notes: 2.0.10 fixes credits and stability

Google Antigravity has a small but useful release notes entry: version 2.0.10, dated May 28, 2026. This is not a major product launch. It is more practical than that: Google lists Various reliability and usability improvements and a G1 credit bug fix. For teams already testing agent work, that is exactly the kind of change worth noting before starting larger runs.

Google Antigravity release notes: what changed in 2.0.10?

Google Antigravity is Google's agentic development platform for launching, guiding and reviewing AI agents that work with projects, files, terminals and browsers. Antigravity 2.0 is the standalone desktop surface for managing those agents across projects and workspaces.

The changelog lists version 2.0.10 as an Antigravity 2.0 release from May 28, 2026. It has two visible items: reliability and usability improvements, plus a G1 credit bug fix. Google also says new versions roll out gradually and may take a few days to reach all users.

Source: Google Antigravity Changelog

Source: Google Antigravity documentation: Home

Why this changelog matters

Short version: if you use Antigravity for real work, treat 2.0.10 as an operational signal, not a marketing story. A coding agent is an AI assistant that can read project files, propose changes and use tools within the access you grant it. When that environment gets a credit fix and stability improvements, it affects how confidently you can plan small, measurable agent runs.

I would not read more into it than the source says. The changelog does not explain exactly how G1 credits are counted or displayed. But if you already track cost, quotas or internal usage, the next sensible test is simple: run one bounded agent task, save the output, check which approvals were needed and compare it with how the same routine behaved before the update.

Short example: use the new Antigravity feature

Human step: When version 2.0.10 has reached your Antigravity 2.0 environment, open a non-critical project where you can already review changes. Then start the agent work there. The source does not verify a new button, menu or shortcut for 2.0.10, so do not build the routine around an invented UI path.

Prompt to the agent:

Before we spend G1 credits on a larger Antigravity 2.0 agent run, inspect this project and propose the smallest useful verification task. List files you need, commands you would run, permissions or approval gates, and what output would prove the task is worth running. Do not change files yet.

Good output should:

• Separate verified files from assumptions
• Treat G1 credits as planning context, not invent a cost model
• Suggest concrete commands or manual checks before anything runs
• Mark where a human should approve file changes, external tools or longer agent runs

What Swedish teams can test today

For small businesses, schools and internal automation teams, the point is not to chase every version line. The point is to build a habit: when the agent platform changes, test one real but small routine. A good first routine is code review, documentation cleanup or comparing two configuration files.

Tie the test to safe integration in practical terms: scoped permissions, environment variables or secret managers for keys, editing in sandboxed folders, approval before file changes and a simple log of what the agent did. That is where Tool Forge thinking becomes concrete: fewer loose chats, more controllable workflow.

The next thing to watch is whether Google adds more detailed release notes about credits, quotas or how Antigravity 2.0 shows agent usage. Then the advice gets more specific. For now, update calmly, measure small and do not confuse a bug fix with a new pricing or governance model.