TechTalks Edinburgh #10 — Keeping AI-Written Code Safe to Change

Dmitriy Fedoryshchev, a lead software engineer, on how to tell whether agent-written code is making a codebase safer or riskier. His case study is modernising a 600-table, 21TB legacy core into microservices — the fifth attempt after four that failed — with a squad that went AI-native on a custom multi-agent system under one rule: only merge code you can explain. He also covers turning the rules the agent kept breaking into checks it cannot skip, built into his own open-source tooling. Tenth edition of the TechTalks Edinburgh series, at CodeBase on Castle Terrace. Doors 18:00 with pizza and drinks, opening talk 18:30, main talk 18:40, Q&A 19:30, pub afterwards. Free, 100 places on Meetup. Partners include CodeBase and Scottish Technology Club. Why go: the most concrete Edinburgh talk this autumn on AI-generated code in a real legacy migration rather than a demo repo. Best for engineers and tech leads deciding how far to let agents into production.