Understand the auth code your AI wrote
A beginner had auth, a Postgres schema and deployment running in a weekend, then realised they couldn't tell whether the server-side permission checks were secure.
Reviewing & testing AI output
Review and testing time sits on top of agent time. Keep tasks small and test the real user flows yourself.
4 items · 4 stories · 0 security incidents
Iterate instead of trusting one-shot AI rewrites
Sharing Shopify's write-up on rebuilding the Shop app with coding agents, the poster points out that Shopify says one-shot LLM rewrites weren't shippable.
Test risky user flows yourself, not just with AI
A builder waited hours on automated AI testing and still ended up restructuring modules and rewriting the rendering layer.
Budget review and testing time for agent tasks
A developer's takeaway after a year: agents don't automatically speed you up, because your review and testing time sits on top of the agent's run time.