AWS: Why some teams saw 10x velocity with same AI tools
AI Engineer · youtube · 2026-08-29
Amazon observed 50 teams using the same coding AI. Half saw <3x deployment velocity gains, while the other half hit 4.5x or 10x. The difference was "Frontier Development": engineers write 1-2% of code, agents run in parallel for hours.
Five habits of high-performing teams:
- Invest in context: Document mental models and prune bloat as models improve.
- Slow down to speed up: Refactor legacy codebases/improve error messages before agent success.
- Feed, don't babysit: Drive agents via docs, not running conversations.
- Fix intent first: Resolve design in documents before generating code.
- Shift testing left: Use local deterministic mocks for fast agent feedback loops.
The new bottleneck is decision speed.
More from coding & agent
- Andrew Ng: Why Software Engineering Fundamentals Remain Critical in the Age of AI Coding Agents — AndrewYNg · 2026-08-29
- LangChain Adds MCP Support in Open Source, Built on FastMCP — LangChain · 2026-08-29
- Dev built his own provider-agnostic artifact hosting after Claude's sharing limits — miihr_ · 2026-08-29
- Open-source 'universal pipe' connects cloud, local and self-hosted LLMs for free — conifer_v11 · 2026-08-29
- Better Models Need Good Design: 4 Levers for Coding Agents — rajistics · 2026-08-29
- LangChain Academy Hosting Live Workshop on Building Deep Agents — LangChain · 2026-08-29