This talk blends inspiration and hands-on guidance to propose a Human-in-the-Loop approach to AI-assisted development. Starting from user stories and acceptance criteria, Amazon KIRO keeps first-class artifacts (specs, development plan, architectural decisions) and evolves them with each change, generating code and tests accordingly. In an end-to-end demo we build a Todo API (FastAPI/Django), set up CI, then apply a change request: KIRO updates specs, proposes code patches, and verifies with tests. We close with a comparison of code-first vs spec-first, common pitfalls, hardening practices, and a practical playbook for Dev/PM/QA teams.

LLMs are fast, but often leave “black boxes.” This talk shows how to keep humans in the loop and start from specifications with Amazon KIRO—from user stories and acceptance criteria to SPEC/PLAN/ARCH, code, tests, and CI. Live demo (FastAPI/Django) with an end-to-end change request.