Amazon Admits Its Flagship AI Coding Tool Isn’t Good Enough for Its Own Workers to Use
Amazon admitted its internal AI coding tool, Kiro, is insufficient for employee use and will now allow access to OpenAI’s Codex and Anthropic’s Claude Code, despite earlier restrictions. The move follows employee frustration and highlights Amazon’s reliance on external AI tools despite heavy investments in competitors like Anthropic and OpenAI.
Amazon has reversed its stance on its in-house AI coding tool, Kiro, after admitting it is not effective enough for its own employees. In a memo to staff, VP of Amazon software builder experience Jim Haughwout announced that Claude Code will be made available immediately, with OpenAI’s Codex following the next week. Both tools will run on Amazon’s Bedrock platform, a cloud-based AI service. The decision comes six months after Amazon pushed employees to use Kiro exclusively, despite investing billions in competitors like Anthropic and OpenAI. Employees had previously expressed frustration over the limitations on using Claude Code, questioning why Amazon would promote external tools while restricting their own use. Amazon claims 83 percent of engineers still rely on Kiro, but the shift signals a concession to employee demands and industry pressure. The company’s earlier push for Kiro had been seen as unusual given its investments in rival AI firms, including Anthropic and OpenAI, which are leading the AI coding race. The reversal also follows recent AI-related outages at Amazon, which the company linked to poorly implemented AI-generated code. Employees had criticized the inconsistency, arguing that customers would question the reliability of tools Amazon itself did not trust. Despite the change, Amazon maintains that Kiro remains the primary tool for most engineers. The move highlights the challenges of developing proprietary AI solutions while competing in a rapidly evolving market dominated by external players.
This content was automatically generated and/or translated by AI. It may contain inaccuracies. Please refer to the original sources for verification.