Why Are Google Engineers Using Claude?

When Google engineers gain access to Claude, the first question that comes to mind is whether the company has lost faith in its own model, Gemini. The expansion announced in September 2026 points instead to engineers being able to try multiple models in their workflows. Gemini remains Google's primary internal model, while Claude is offered through Antigravity as an additional option for specific coding tasks. That distinction matters: developing a product of its own does not mean a company has to make its engineers use only that product.

Why do Google engineers have access to Claude?

According to Business Insider and TechRepublic, Google expanded access to Claude in September 2026, after it had previously been available only to limited groups. Claude Opus 5 is available through the company's Antigravity development environment, but access is subject to usage limits. A statement from a Google spokesperson also makes the approach clear: Alongside Gemini, engineers can use some third-party models for specialized tasks.

Behind this decision is the fact that software teams care more about the results they get in their day-to-day work than about a model's brand. One model may be better suited to understanding an existing codebase, while another may be better at following a long change plan. Tasks such as debugging, test generation, and code review also call for different skills. For a team running software in production, the question is not “which model won?” but which tool creates less friction for a particular task.

One important caveat: The sources do not show that Google engineers consider Claude better than Gemini. Google has made Claude available as an additional, specialized option. So interpreting this decision as Gemini's failure or as Google changing its model preference would go beyond what the available information supports.

How are Google engineers using Claude?

The Antigravity integration brings Claude closer to the development workflow, rather than leaving it as a general-purpose chatbot in a separate chat window. While working on code, an engineer can get suggestions for changes with help from a different model. The practical benefit is that they don't have to step away from the editor to switch to a new tool. Of course, integration does not mean the code suggested by the model is automatically correct or safe; any generated changes still need to be tested and reviewed.

Putting a usage limit on Claude is no minor operational detail, either. Usage limits can help manage cost and capacity while measuring where the model delivers value. For example, a team might evaluate Claude for code reviews that require long context, and Gemini for other tasks tied to internal company documentation. This kind of choice only becomes meaningful when the models are compared on real tasks. A one-off impressive demo doesn't, by itself, reveal the effect on speed or error rates in a daily workflow.

One thing most sources overlook is that using a third-party model within an organization involves more than simply granting access. Which service receives the code, how sensitive data is protected, how usage records are kept, and which outputs developers can trust are all part of the design. A controlled environment such as Antigravity can make this experience easier to manage; teams still need to check their company policies and data classification rules.

What does the Antigravity integration change?

The real shift is that competition is no longer limited to consumer chatbots. A coding assistant interacts with the developer's editor, repository, testing process, and review habits. Once a model becomes part of this workflow, its value is measured not only by the quality of its answers but also by how seamlessly it lets developers keep working. That's why the race between Gemini and Claude is becoming visible in real development environments, rather than just in product announcements.

Reports by the Los Angeles Times in April 2026 about competition over AI coding within Google put this decision in a broader context: Companies were said to be feeling pressure to improve developer productivity, tools from competitors such as Anthropic were attracting attention, and Google was reportedly trying to bring together different coding initiatives. The expansion of access in September may not be the only response to this pressure, but it does give Google a chance to track developers' preferences more closely.

News reports of Google's plan to invest up to $40 billion in Anthropic also show that the relationship between the two companies doesn't fit a simple “competitors keep their distance” pattern. Investment, collaboration, or product use does not eliminate competition between models. Large technology companies can be competitors while also using one another's tools to meet specific needs.

What can teams learn from the AI race?

Google's choice shouldn't be treated as a formula for choosing a model for engineering teams. A more useful takeaway is to make tool decisions task by task. If your team has two coding assistants, give both the same real tasks: finding a bug in an existing module, adding tests, writing an explanation for a change, and producing a solution that fits the codebase. Then measure not only the code they produce but also the time spent fixing it, failed tests, and how much developers change the suggestions.

This approach turns model selection from a debate about personal preference into an observable engineering decision. Keep data within your organization's security rules, account for usage costs, and leave developers responsible for reviewing every suggestion. If one model clearly performs better on certain tasks, use it for those tasks; if you can't measure an advantage, don't make a new tool the default just because it's popular.

What Google's example points to is controlled use of multiple models rather than reliance on a single model. Keeping Gemini as the default while providing access to Claude subject to usage limits is not a contradiction. A concrete step your team can take today is to compare the two tools in a small pilot using the same task, and record which one actually saves time for each type of work.

Why do Google engineers have access to Claude?

How are Google engineers using Claude?

What does the Antigravity integration change?

What can teams learn from the AI race?

Sources