Google, one of the world’s largest AI companies, quietly did something unexpected on September 15: it opened Anthropic’s Claude Opus 5 to every engineer at the company through its internal development platform, Antigravity. For a business that built and heavily promotes its own Gemini models, this is a meaningful admission about the state of AI tooling in 2026.

What Changed

For most of its existence, Google kept a tight policy around external AI tools. Most employees were barred from using Claude Code or OpenAI’s Codex for work tasks, pushed instead to build with Gemini. Exceptions existed for teams at Google DeepMind and for specific high-priority engineering projects, but broad access was off the table.

That changed. Google confirmed that Claude Opus 5 is now available to all engineers through Antigravity, its internal coding and agent orchestration platform. Access comes with a usage quota per employee, positioning Claude as a complement to Gemini rather than a replacement. The company statement made the framing explicit: “Gemini remains our primary and foundational model for internal development.”

In other words, Gemini is still the official line. But for the coding tasks where Claude outperforms, engineers can now use it.

Why This Matters More Than It Looks

It would be easy to frame this as a minor internal IT change. It is not.

Google is an investor in Anthropic, which already creates a somewhat unusual dynamic. But more than that, Google is in the business of convincing enterprises to build on Gemini. Giving engineers access to a rival model signals something the company may not want to say out loud: different AI models are genuinely better at different things, and picking only one costs you performance.

That is exactly what most enterprise AI buyers are discovering in 2026. The early “we will standardise on one model” approach is giving way to a more practical question: which AI is actually best for this specific task?

The fact that Google’s own engineers are now using Claude for coding work is a data point that resonates beyond the walls of Mountain View.

The Antigravity Layer Is the Real Story

What Google is doing with Antigravity is also worth paying attention to. The platform acts as an orchestration layer that can route work to different models depending on the task. It supports Claude alongside Gemini, rather than presenting one as the universal answer.

This architecture reflects a broader trend in enterprise AI: the model layer is increasingly commoditised, and the value sits in the orchestration, governance, and integration layer above it. Companies that lock users into a single model are fighting a losing battle, because the best model keeps changing.

For businesses building their own AI infrastructure, Google’s internal approach is a quiet endorsement of the multi-model strategy: use Gemini where it leads, Claude where it leads, and build a platform that routes intelligently between them.

What This Means for Business

The “best AI” conversation has shifted. A year ago, most enterprise AI buyers were asking which single platform to commit to. Today, the more useful question is which AI is strongest for which category of work. Google’s engineers are now living that reality.

Claude’s coding capability is proven at the highest level. When the company that employs some of the best engineers in the world and builds frontier AI models decides its own engineers benefit from access to a competitor’s coding tool, that is a meaningful signal. It is not marketing. It is a practical decision made by people who have tried both.

Multi-model orchestration is becoming standard. The ability to route tasks to the right model, with governance and quota controls in place, is precisely what enterprise AI platforms need to provide. Antigravity is one implementation. The pattern will spread.

AI model loyalty is for consumers, not enterprises. For personal use, picking one AI assistant and sticking with it is fine. For business operations, the more important question is whether your AI infrastructure can use the best model for each job. The companies building that flexibility into their systems now will have a meaningful advantage in 18 months.

---

The move also raises a practical question for every business not named Google: if you are still running all AI work through a single model because it is easier to manage, are you leaving meaningful performance on the table?

The answer, based on how Google is now running its own engineering, is probably yes.