before AI slop, there was human slop Singaporean Cursor engineer @poteto, shipping ~1,000 PRs a month with coding agents, says it comes down to one thing: “Let the agent run and test its own code. It doesn't guarantee good code, but it allows correct code ... I've gotten to a point where I really don't look at the code anymore.” most people give agents a prompt and pray She gives her agents the ability to run, test, and verify their own work, closing the feedback loop and shipping at a scale most developers can only imagine Prompt → Agent writes → Agent runs → Agent verifies → Ship The real skill isn't getting AI to write more code. It's getting AI to verify its own work a few minutes that might completely change how you use coding agents Bookmark this. Watch it, then learn how to build agents that don't fall apart ↓