Amazon engineers are openly pushing back against internal policies that heavily favor the company’s own AI coding assistant, Kiro, with many arguing that Anthropic’s Claude Code delivers significantly better results for complex development work.
According to a detailed report by Business Insider published on February 11, 2026, Amazon has been actively steering developers toward Kiro for production code and live systems, while requiring formal approval processes for third-party tools such as Claude Code. This has sparked frustration across engineering teams, especially since Amazon is one of Anthropic’s largest investors (with roughly $8 billion committed) and sells Claude models to customers through AWS Bedrock.
In one major internal forum thread, nearly 1,500 Amazon employees endorsed a request to formally approve Claude Code for broader internal use. Some engineers pointed out the irony of being expected to recommend Claude Code to AWS customers while being discouraged from using it themselves for production work. Others argued that restricting tool choice is hurting developer productivity and velocity.
Amazon maintains there is no outright ban on Claude Code. However, the company’s internal guidance strongly recommends Kiro as the default tool for production environments, citing better performance, tighter security controls, and faster iteration cycles. An Amazon spokesperson told Business Insider:
“We are seeing incredible improvements in efficiency and delivery from Kiro… and we want to make sure our internal employees all take advantage of this capability to deliver faster for our customers.”
Kiro, launched in mid-2025, is Amazon’s homegrown AI coding agent built on top of Anthropic’s Claude models combined with Amazon’s proprietary tooling and infrastructure.
The situation highlights a growing tension inside Big Tech: companies investing billions in frontier AI labs while simultaneously trying to force adoption of their own branded tools internally, a classic “eat your own dog food” strategy that isn’t always popular when superior alternatives exist.
This story is part of a broader industry pattern. Similar debates have surfaced at Google (Gemini vs. internal tools), Microsoft (Copilot vs. third-party agents), and Meta, as engineering teams increasingly prefer the best-performing model regardless of corporate ownership.
As frontier coding agents continue to improve rapidly, the pressure on large tech companies to allow their engineers to use the best available tools, rather than mandated internal products, is only expected to grow.
