Six engineers. Seventy-six days. That’s what it took AWS to rebuild the inference engine behind Amazon Bedrock, its flagship AI service. The original estimate called for 30 to 40 engineers working 12 to 18 months.
The result is a new architecture called Mantle, or Project Mantle, designed to solve the capacity and performance bottlenecks that emerged as AI workloads on Bedrock exploded.
How six people did the work of forty
The secret ingredient was Kiro, Amazon’s agentic coding service. The tool generated specs, code, tests, and deployments, essentially acting as a force multiplier that turned a small senior team into something far more productive. AWS leadership has cited productivity gains of 10 to 20 times when skilled engineers harness agentic AI tools within controlled workflows.
Andy Jassy referenced the Mantle project in his 2025 shareholder letter, positioning it as evidence that Amazon practices what it preaches when it comes to AI adoption.
Despite the heavy use of AI-generated code, AWS leadership has stressed that mandatory human review remains a non-negotiable part of the process. Every line of AI-produced code gets checked by humans before deployment.
Why Bedrock needed a rebuild
The timing of Mantle wasn’t arbitrary. Bedrock processed more tokens in the first quarter of 2026 than it did in all previous years combined. Service usage nearly doubled in March 2026 alone.
The Mantle engine now operates through a dedicated bedrock-mantle service endpoint, separate from the legacy bedrock-runtime endpoint. AWS documentation confirms the separation of quotas between the two, indicating the new architecture is fully live and serving production traffic.
The competitive landscape
AWS has been in a well-documented fight for AI market share, particularly against Microsoft Azure, which has benefited enormously from its partnership with OpenAI. Google Cloud has also been aggressive, leveraging its Gemini models and TPU hardware.
The Mantle project also validates a specific approach to AI-assisted development: small, senior teams using agentic tools with rigorous human oversight.