← Back to feed

Amazon has formally opened Anthropic’s Claude Code and OpenAI’s Codex to its entire corporate engineering workforce, ending months of internal restrictions that had frustrated developers and created an awkward contradiction for the company’s sales teams.

In a note to staff obtained by Business Insider, VP of Amazon Software Builder Experience Jim Haughwout confirmed Claude Code is available company-wide immediately. Codex follows May 12. Both run on Amazon Bedrock, keeping all inference traffic inside AWS infrastructure.

The Numbers

• Claude Code: company-wide access, effective immediately
• Codex: follows May 12, same Bedrock-managed path
• Kiro: still primary; 83% of Amazon engineers use it as their main tool
• 1,500+: number of Amazon engineers who formally endorsed Claude Code adoption in internal forums before the reversal

Why It Took This Long

Until now, Claude Code was not approved for production use without formal clearance. The restriction forced engineers to route exceptions through a separate approval process while the tools remained officially unsupported. Internal discussions surfaced an obvious tension: Amazon sales engineers helping enterprise customers deploy Claude Code on Bedrock were operating a tool they were not permitted to use themselves in production.

The policy created friction across engineering teams. A manager overseeing internal developer tools acknowledged the “frustration” and noted the roughly 1,500 internal endorsements calling for formal adoption. Some engineers argued Claude Code outperformed Kiro on specific workloads and that forcing adoption of a less capable tool was slowing delivery.

A previous internal guideline had indicated Claude Code cleared security and legal review — that language was later edited out, deepening the confusion.

The Infrastructure Rationale

Running both tools through Bedrock solves the enterprise compliance problem that made the previous policy untenable. Inference stays within Amazon’s own cloud perimeter. Teams do not need to provision separate capacity, manage API keys externally, or configure data handling separately. From a security posture, the arrangement is cleaner than ad-hoc developer accounts on third-party APIs.

“Both run on Bedrock, where all inference runs,” Haughwout wrote. “Both will have easy install for all Amazon builders.”

Amazon described the move as “standardising” access rather than abandoning Kiro. The 83% Kiro adoption figure suggests most engineers default to the in-house tool. Claude Code and Codex now sit alongside it as officially sanctioned options — the company’s language is explicitly “additional tools,” not replacements.

What This Signals

Amazon has committed $25 billion to Anthropic in a multi-phase investment, with Anthropic agreeing to purchase $100 billion in Trainium capacity from AWS. A separate Amazon-OpenAI partnership involves up to $50 billion in exchange for OpenAI using AWS infrastructure. In both cases, Amazon has a commercial incentive for its engineers to understand these tools from the inside.

The rollout removes a visible credibility gap: a cloud platform selling AI coding tools it did not formally permit its own staff to use. More broadly, it reflects where enterprise AI tooling has landed in 2026 — not as optional productivity experiments, but as infrastructure decisions that require the same governance treatment as any other cloud service.

Kiro being used by 83% of engineers means the rollout does not displace Amazon’s internal tooling investment. It extends the envelope. Whether that changes over time as engineers accumulate direct experience with all three tools is the data that will matter in future internal reviews.