They promised us that coding agents would be pure magic for coding. Reality? We’re back to the fundamentals of software craftsmanship - just with a new, sharp tool in our hands.
In this talk I’ll share real-world stories from adding coding agents on top of legacy Android/iOS codebases: where they helped, where they quietly made things worse, and what we’ve broken along the way.
Instead of shiny „best practices” (they don’t exist yet), you’ll get a field report of worst practices: how we misused agents for refactors, tests, and documentation, what kinds of hallucinations and architectural chaos they created, and how we had to fall back on boring things like boundaries, small steps, and good review habits.
You’ll leave with a practical checklist for using agents in your own projects: where to start safely, what NOT to delegate, and how to keep AI from turning your codebase into an unmaintainable black box.