Amazon Expands Developer Toolset: Claude Code and Codex Join Kiro on AWS

Developers collaborating with AI assistants (Kiro, Claude Code, Codex)

Amazon has quietly shifted the rules of engagement for its internal developer community. In a recent staff note, the company announced that tens of thousands of its developers will now have immediate access to Anthropic’s Claude Code and, soon, OpenAI’s Codex — both hosted on AWS and Amazon Bedrock. The move signals a notable loosening of earlier restrictions that favored Amazon’s own agentic coding service, Kiro, and underscores a broader industry turn toward making third-party AI coding assistants an accepted part of production workflows.

What changed inside Amazon

Amazon previously encouraged use of Kiro, its Bedrock-based agentic coding platform launched in 2025, for production code. Internal guidance reportedly pushed teams to prioritize Kiro over non-approved third-party tools, which caused frustration among engineers who wanted the flexibility to use the best assistant for the task. After vocal internal pressure — including an employee thread with roughly 1,500 endorsements for Claude Code — leadership relented. The company approved Claude Code for production use immediately and announced Codex will follow, removing the need for special clearance to use these tools. Kiro remains available, but the policy shift reflects a more permissive stance toward external AI tooling, provided it runs within Amazon’s controlled AWS environment.

Why developers pushed for third-party tools

Engineers often choose tools based on the fit for a specific problem, personal workflow preferences, and perceived productivity gains. Claude Code and Codex each offer different model behaviors, prompt interfaces, and strengths, and many developers had already adopted them informally. The internal advocacy at Amazon highlights a recurring dynamic in large engineering organizations: a tension between central platform standardization (for security, supportability, and observability) and developer choice (for speed, creativity, and effectiveness). By enabling third-party agents on Bedrock, Amazon aims to balance both priorities — offering choice while centralizing infrastructure, capacity management, and compliance.

Impacts on development workflows

The practical effects for engineers are immediate and meaningful. With multiple agentic assistants available, teams will likely shift how they partition labor between human engineers and AI: less time spent typing boilerplate and more time on architecture, integration, and validation. That reallocation brings new needs — stronger review processes, clearer validation checks, and tooling to catch subtle behavioral mismatches between agent outputs and system expectations. Leaders will need to rethink productivity metrics, since output may look different when agents generate much of the initial code. The new status quo nudges organizations toward defining guardrails for agentic behavior, test harnesses that validate AI-generated code, and governance that distinguishes exploratory usage from production-grade delivery.

Broader implications for e-commerce and AI adoption

Amazon’s move is part of a broader pattern in retail and e-commerce where AI, especially agentic systems, is accelerating from experimentation to embedded capability. Retail peers like Walmart have already reported substantial developer-hour savings from AI coding assistants and have built custom agentic tools for retail-specific tasks. For Amazon, widening access to Claude Code and Codex could speed feature development for retail services (e.g., Rufus, its shopping assistant) and internal platforms alike. It also reinforces the notion that AI-enabled coding is becoming a baseline expectation rather than a niche productivity experiment.

Operational and security considerations

A central technical rationale for Amazon’s decision is operational control: running third-party models on AWS Bedrock lets the company avoid complex external integrations, simplifies capacity planning, and enforces data security and compliance. Even as developers gain choice, the hosting constraint ensures that data residency, telemetry, and access controls remain consistent with corporate policy. Still, organizations will need to extend security reviews to agent prompts, model outputs, and downstream dependencies to prevent leakage of sensitive information or the introduction of insecure patterns into production code.

Strategic partnerships and what’s next

The update comes alongside deepening commercial ties between Amazon and major AI vendors. Anthropic’s Claude platform availability on AWS and Amazon’s multi-billion-dollar investments in both Anthropic and OpenAI show a strategic alignment: Amazon benefits by offering best-in-class models on its infrastructure, while partners gain scale and distribution. For developers, that means a richer toolset delivered under a managed umbrella, but it also raises questions about vendor lock-in, cost management, and how teams will benchmark across competing models over time.

What to watch

Governance frameworks: How will Amazon—and similar enterprises—codify acceptable uses, validation steps, and audit trails for agentic outputs?
Productivity measurement: Will organizations adopt new metrics that capture value from AI-assisted development rather than raw lines of code?
Interoperability and portability: Can organizations standardize interfaces so switching between Kiro, Claude Code, or Codex is seamless when needed?
Security posture: How effectively teams instrument and monitor AI-generated code for vulnerabilities and compliance drift?
Competitive dynamics: How will other large tech and retail companies respond in policy and tooling, especially where custom agentic systems are already in use?

Conclusion

Allowing Claude Code and Codex to run on AWS alongside Kiro is a pragmatic compromise: it gives engineers access to the assistants they want while keeping control firmly within Amazon’s infrastructure. The change is emblematic of a wider transition in software development — one where AI agents become part of the standard toolkit, reshaping roles, workflows, and the organizational practices needed to manage them safely and productively. For engineering leaders and practitioners alike, the priority now is to build the guardrails, review processes, and measurement systems that let agentic coding accelerate innovation without sacrificing quality or security.