Anthony LiguoriVP and Distinguished EngineerAmazon Web Services
An AWS team of engineers at L7 or above, working on Amazon Bedrock in Kiro's IDE and CLI.
Kiro (AWS), published Read the case study (opens Kiro (AWS) in a new tab)
Observation
Six senior AWS engineers rebuilt part of Bedrock's inference engine with Kiro, far inside AWS's own estimate.
- Takeaway
- Fix the test and deploy loop and size each task to one commit before expecting agent speed. This team reviewed every AI-written commit itself.
- What happened
- AWS reports that six senior engineers shipped about 100,000 lines of Rust in 76 days, against an estimate of 30 engineers for 12 to 18 months. They wrote about 95% through Kiro. Each prompt produced one reviewable commit, and modules stayed small enough for model context. Senior engineers reviewed every AI-written commit themselves, replacing the traditional review process. The first weeks went into redesigning the workflow.
- Vendor case study
- AWS's own team, vendor-reported outcomes
- No model named
Deployment details
An AWS team of engineers at L7 or above, working on Amazon Bedrock in Kiro's IDE and CLI.
- Vendor case study
- AWS's own team, vendor-reported outcomes
- No model named
- Team
- Six senior AWS engineers, later 11
- Duration
- 76 days to production
- Workload
- Rebuild of part of the Amazon Bedrock inference engine, about 100,000 lines of Rust
- Product version
- Kiro IDE and CLI with specs, steering, and hooks
- Measurement
- Commits, lines shipped, and days to production as reported by AWS; the 20x productivity figure is the team's own estimate
