Google just admitted its own billion-dollar AI coding tool can't keep up — and gave its engineers access to a rival's model that Google itself is bankrolling.

The move confirms what a head-to-head test already showed: when it comes to AI coding assistants, a handful of Big Tech firms are tightening their grip on the infrastructure of all future software development. Whoever controls the code controls what gets built — and right now, that power is concentrating fast.

XDA Developers ran the three flagship AI coding agents against the same prompt: rebuild a complex food-delivery website with animations, filters, responsive layouts, and interactive elements. No hand-holding, no follow-up hints. Claude Code, running Anthropic's Opus 5, took the crown. Google's Antigravity, powered by Gemini 3.8, earned second place — strong overall presentation, smooth animations, but laughable mobile mockups and a cramped FAQ section. OpenAI's Codex, running GPT-5.6, made a solid first impression but fell apart on inspection: small typography, a rushed newsletter section, and the same pattern of needing follow-up corrections that users have reported before.

Then came the real tell. Business Insider reported that Google has now made Anthropic's Opus 5 model available to all engineers through its internal Antigravity development platform — a sharp reversal from the company's previous policy of barring outside coding tools and pushing staff to build with Gemini. Google DeepMind and select high-priority teams already had Claude access; now the whole company gets it, with per-user quotas, described as "supplementary" to Gemini.

The frustration inside Google was real. Despite recent gains from its Flash 3.8 model, Google is still trailing Anthropic and OpenAI on coding, and employees felt the squeeze — pressured to show AI-driven productivity gains while being denied the best tool for the job. Amazon caved to the same pressure earlier this year, allowing its engineers to use Claude and Codex.

Now follow the money. Google announced plans earlier this year to invest up to $40 billion in Anthropic. So when Gemini falls short, engineers switch to Claude — and Google still profits from the rival's success. A Google spokesperson framed it as routine: "Engineers have access to select third-party models in Antigravity, which is aligned with our external Antigravity enterprise offering. Gemini remains our primary and foundational model for internal development."

That's not competition. That's consolidation. The same handful of firms fund each other, license each other's models, and lock out independent developers who can't match their compute or capital. The open-source alternatives that could break this stranglehold are still playing catch-up. The question isn't which AI writes better code. It's whether anyone outside this club will ever get a seat at the table — or whether every line of code in America will someday need permission from three companies to run.