Here is the most embarrassing detail from the LA Times' devastating look inside Google's AI coding crisis, published Tuesday: some Google engineers prefer to use Anthropic's Claude Code over Google's own tools. Not just prefer. Some DeepMind teams, including those working on the Gemini model itself, actively use Claude Code with management exceptions.

Google is building AI for the world while its own people think the competition's product is better. That is the kind of problem money cannot fix.

Six Products, Zero Focus

The core problem, according to current and former employees who spoke to the LA Times, is fragmentation. Google's Gemini model capabilities are scattered across half a dozen coding products with different branding. Google Cloud, Google Core, Google Labs, Android, and DeepMind are all pushing AI coding initiatives in different directions. Nobody is in charge. Or rather, everyone is.

Anthropic has Claude Code. OpenAI has Codex. These are singular, branded products with clear identities. Google has Antigravity (internal), Agent Smith (internal tool for automating coding and documentation), Jules, Project IDX, Firebase Genkit, and whatever DeepMind is cooking separately. Ask a developer which Google AI coding tool to use and they will stare at you.

Chief AI Architect Koray Kavukcuoglu is now working to unite Google's internal AI coding tools under Antigravity, the platform released last year. But the LA Times reports that confusion about who leads AI coding persists. DeepMind has formed a new team led by research engineer Sebastian Borgeaud. Nobel Prize winner John Jumper is also working on AI coding. These are serious names doing serious work. But serious names in competing silos is exactly how Google lost search advertising to nobody (because it had a monopoly) and AI coding to a startup called Cursor and a safety lab called Anthropic.

Why Coding Matters More Than Anything Else Right Now

Keith Zhai, co-founder of startup TinyFish, told the LA Times something that every AI executive should tape to their monitor: 'Coding is the single easiest way to actually make money.' He added that engineers in Silicon Valley toggle between Claude Code and OpenAI's Codex constantly, but Google is 'often not even in the conversation.'

Raj Gajwani, former Google executive and now CBO at OpenArt AI, laid out the strategic stakes: 'If you win at coding this year, you get the raw data you need to win at model capability next year.' AI coding is not just a product category. It is the flywheel that makes everything else better. The models that write the most code generate the most feedback data, which trains the next model, which writes even more code. Anthropic claims AI assists on nearly all of its own coding. Google has 75% AI-generated new code, but the internal tools producing that code are apparently not good enough for its own engineers to prefer them over Claude.

The Ban That Does Not Work

Most Google employees are banned from using competing tools like Claude Code or Codex due to security concerns. But they can request exceptions with a demonstrated business case. And some clearly have. The LA Times reports that DeepMind teams working on Gemini, internal applications, and open source models use Claude Code. 'You want the best people to use the best tool, even inside Google,' one former employee said.

This is ironic on multiple levels. Google has invested billions of dollars in Anthropic. Anthropic cut off OpenAI's access to its models last year, according to Wired. But Google's own engineers, building the products that compete with Claude, are quietly using Claude to do it.

The Strike Team

Enter Sergey Brin. Google's co-founder has been pushing to close the AI coding gap, with engineers outside DeepMind reportedly being sent to mandatory AI training sessions. The 'Coding Strike Team' we reported on earlier this month is real, assembled specifically because Google leadership sees Anthropic pulling ahead with Claude Code in the enterprise market.

A Google spokesperson told the LA Times the company has seen 'tremendous adoption' of internal tools like Antigravity. And that is probably true. But tremendous adoption of tools your engineers would replace with a competitor's product if they could is not the same as having the best product.

The Timing Problem

Google has the models. Gemini 3 was widely viewed as the best AI model available when it launched late last year. Google has the money. It has the talent. It has the distribution. What it does not have is the focus that comes from being a startup that does one thing well.

And in the AI coding market, even a momentary lag is consequential. While Google reorganizes internally, Cursor just hit $50 billion in valuation. Anthropic's enterprise AI spending share passed OpenAI's last quarter. The coding wars are being won by companies with fewer people, less compute, and a fraction of Google's budget. They just have one product that works and engineers who actually want to use it.

Google Cloud Next 2026 is happening in Las Vegas right now. Google is announcing massive platform plays and next-gen silicon. But somewhere in those Vegas hallways, a Google engineer is probably using Claude Code to prepare their demo. And that tells you everything.