Your AI agent isn't dumb. Your brief is. I hand work to AI agents every day — … · @ReignSwords · TwKit

@ReignSwords

Your AI agent isn't dumb. Your brief is. I hand work to AI agents every day — browser tasks, research runs, coding agents. The single biggest quality jump I ever got wasn't a better model. It was a better brief. Same agent. Same model. A one-line brief came back with the wrong metric from the wrong page. The same job with a five-line brief nailed it on the first try. Nothing changed but the words I gave it. Three lessons, all paid for in failed runs: 1. The agent only sees what's in the brief. I once attached reference files to a task — the agent couldn't open them. Three failed runs before I pasted the 58 IDs directly into the brief body. Attached-and-prayed files are decoration. 2. Vague instructions leak. I told a running agent to "just report the status." It read that as "wrap up now" and submitted a half-done report. Agents don't do vibes. They do stop rules. 3. "Done" must be verifiable. "Check the new followers" is a wish. "List new blue-check followers since Tuesday, verify each on their profile page, report handle + badge status" is a task. My five-line brief, every time: Goal. Inputs pasted inline. Done stated as checkable facts. What NOT to do. Report format. Write that, and a weak model with a sharp brief beats a frontier model with a lazy one. I watch it happen every week. Fellow builders — when your agent flopped last time, be honest: was it the model, or was it your brief? I'm still finding my own bugs. 🔵 Fellow Premium members, let's connect! I post about AI & tech — follow for more 👥

曝光 122 · 评论 7 · 点赞 2 · 书签 0