Amazon Engineers Push Back on Internal AI Tool Rules

Amazon’s internal teams are feeling the squeeze from new rules that limit their use of Anthropic’s Claude Code in everyday work. Even though the company backs Anthropic heavily and sells its AI services through AWS Bedrock, employees can’t deploy Claude in live products without special permission. Late last year, Amazon rolled out guidance that nudges teams toward its own coding assistant, Kiro. The memo urges developers to use Kiro for production code and lists Claude as a non‑approved third‑party option. The move sparked heated debate inside Amazon’s forums. Roughly 1, 500 staffers signed a thread asking for formal adoption of Claude Code, arguing that the tool offers better performance than Kiro. The policy hits hard on engineers who sell AWS Bedrock, Amazon’s platform that lets customers access Claude and other AI services. One engineer wrote that it is hard to convince clients of a tool you can’t use yourself in internal projects. Amazon’s relationship with Anthropic is layered: the tech giant invests over $60 billion in the startup, while Anthropic relies on Amazon’s cloud and chips. The company claims it wants to boost internal efficiency with Kiro, but says it will not open‑source or fully support other AI tools for production.A sales engineer highlighted the irony of selling Claude through Bedrock while it’s not officially allowed inside Amazon. He noted that to support customers, he needs to demo and build with Claude himself. Other engineers worry Kiro’s slower performance could hurt development speed, calling it a “survival mechanism” rather than a genuine innovation. About 70 % of Amazon’s software engineers used Kiro at least once in January, yet many still favor Claude. A manager overseeing internal tools acknowledged the frustration and cited 1, 500 endorsements for Claude’s official use. Some employees pointed out that earlier internal guidelines had cleared Claude for production, only to later remove that language. The lack of clear decision‑making data left many questioning why the tool was denied official status. The overall picture shows a company juggling its own AI ambitions with partnership obligations, while employees push for tools that they believe will deliver better results.