Amazon rolls out Codex and Claude

AI coding assistants are turning into standard workplace infrastructure inside big tech — not side experiments. That’s the real story here. Amazon has decided to give its corporate employees broad access to Anthropic’s Claude Code right away, with OpenAI’s Codex set to follow on May 12. The move matters because Amazon had been steering people toward its own tool, Kiro, while limiting access to outside options. Now that wall is coming down. ### What changed inside Amazon? Amazon is rolling out two outside coding agents across its corporate workforce instead of keeping them in a narrow pilot or approval-only lane. Claude Code is available company-wide immediately, and Codex is scheduled for May 12. Both will run on Amazon’s own stack, which means employees get the tools without every team building separate plumbing first. ### Why is that a bigger deal than it sounds? Because this is not just “more tools.” It reverses a pretty awkward internal setup. Amazon had invested heavily in Anthropic and also worked with OpenAI through AWS, but many employees still couldn’t easily use the strongest coding assistant boxed into a narrower menu. ### Why were employees pushing so hard? Turns out the pressure had been building for months. Engineers argued that limiting Claude Code in particular was slowing them down and making Amazon look behind the curve on developer productivity. In one internal discussion, about 1,500 employees endorsed power users — it had real internal weight. ### Where does Kiro fit now? Kiro doesn’t disappear. But it stops being the only politically safe answer. Amazon had previously nudged teams toward Kiro for production work, which made sense if leadership wanted tighter control and a homegrown default. The catch is this: If it doesn’t, they route around it. ### Why run Claude Code and Codex through Bedrock? Basically, Amazon gets two wins at once. Employees get easier access, and Amazon keeps the experience inside AWS-managed infrastructure. That means less setup pain, more centralized control, and a clear path to use. ### Does this say anything broader about enterprise AI? Yes — and it’s pretty simple. Big companies are learning that “pick one vendor and force everyone onto it” is not how AI tooling settles in. Coding agents are becoming more like cloud services or IDEs. Teams want options. Management sees that pattern almost perfectly. This last point is an inference from the rollout details and the employee pushback, but it’s the cleanest way to read what changed. ### So