WWithAviinwithavi.hashnode.dev·2d ago · 4 min read20 cases before launch: a minimal eval setYou do not need a platform to start testing an AI feature. You need about twenty real cases, a rule for passing, and a rule for when not to ship. Plenty of teams launch an AI feature after trying it o00
WWithAviinwithavi.hashnode.dev·2d ago · 4 min readHow to write a task for a coding agentA clear task means less guessing for the agent. It does not promise correct code, so say how you will check the result too. Give a coding agent a fuzzy task and it may quietly fill the gaps with assu00