Reports published on September 16, 2026 say Google has expanded internal access to Anthropic’s Claude for engineers through its Antigravity tool. The report does not mean Google has replaced Gemini. Instead, it shows why large engineering organizations may use more than one model when coding tasks have different strengths and constraints.
What is being reported
Business Insider reported that Google is allowing all engineers access to Claude through its internal Antigravity environment, with a quota system. The report says Google had previously restricted broader access while promoting its own Gemini models.
Why multi-model workflows are becoming normal
Teams do not always need the same model for every job. One model may fit a fast coding task, another may fit a long code review, and another may integrate better with a company’s existing platform. The useful question is therefore about the workflow, not the brand name alone.
Evaluate the complete coding workflow
When comparing coding assistants, measure the task from start to finish. Look at repository understanding, edit quality, test generation, tool use, error recovery, review time and the controls around production actions. ToolBoxKart’s AI agent software testing workflow is a useful companion for this kind of evaluation.
| Area | What to check |
|---|---|
| Codebase context | Can the tool find the right files? |
| Changes | Are edits narrow and reviewable? |
| Testing | Does it run useful checks? |
| Security | Are permissions and secrets controlled? |
Do not compare models only by benchmark numbers
A benchmark can help, but coding productivity also depends on the tools around the model. Context handling, repository access, test execution, approvals and integration quality can change the result. Run the same real tasks through each workflow before standardizing.
What this means for developers
Developers can benefit from a small approved set of coding tools instead of forcing one model into every job. Set clear rules for where each tool can be used, keep credentials protected and make the final production step reviewable.
Keep the security boundary clear
Giving a coding assistant access to a repository is different from giving it access to production credentials. Use separate environments and least-privilege permissions. Keep secrets out of source files, prompts and generated artifacts.
FAQ
Does this mean Google is abandoning Gemini?
The cited report does not say that. It describes broader access to Claude for engineers while Google continues to develop and use Gemini.
Should every engineering team use multiple models?
There is no universal requirement. Teams should test the workflows they actually use and choose a setup that fits their tasks, controls and costs.
Sources
- Business Insider — Google finally lets all engineers use Anthropic's Claude
- OpenAI — The next evolution of the Agents SDK