Q&A
If my AI output is sloppy, is it the AI's fault or mine?
Short answer
Almost always your own instructions. Treat AI as your team and yourself as the manager. When a direct report hands in sloppy work, the brief is usually vague: no clear goal, no example, no definition of done. The model can be wrong too, but it is the second suspect, not the first.
Your AI output is bad. What is your first reaction? Most people blame the model. I did too.
Try the manager's view instead. On Monday you tell a new hire, "Make me a report." No deadline. No example. You don't say who will read it. On Friday the report lands, and it's sloppy.
Do you blame them? An honest manager says: my brief was bad.
AI works the same way. One difference: it won't walk back to your desk and ask "who is this for?" unless you tell it to. It just starts, using its best guess. That guess is the sloppy result you're looking at.
These are the five things that used to be missing from my briefs:
- Goal. What decision is this report supposed to support?
- Context. Who reads it, and what do they already know?
- Examples. What does good look like to you?
- Constraints. What must not be touched or changed?
- Definition of done. When do you call it finished?
I felt this directly while building a content distribution system across 8 channels, which I run alone. It only became stable when I stopped typing "please make me a post" and started writing a brief per channel, the way I'd assign work to a real team. Same model. The only thing that changed was how I handed over the work.
Review works the same way. When I built a customer support agent, a data access gap between users got caught before production. The model didn't magically hold itself back. I told it to check, and I read the results myself. A manager stays accountable for what the team ships.
I'm not saying models never fail. Sometimes the data really isn't there, or the information is stale. But there's an easy test. Hand your brief to a smart person who just joined the company. If they'd be confused, the brief is the problem. If they'd understand it and could do the work, now you can suspect the model.
Making companies leaner is an old habit of mine, from well before AI was trendy. As a product leader I drove $4 million+ in savings per year across a range of initiatives: process improvements, cost optimization, and more efficient product decisions. Automation was one of them, for example taking auto-resolved customer support tickets from 0 to 70% of roughly 10,000 monthly tickets. Back then, that kind of leverage took a senior seat, an engineering team, and expensive systems. Today AI is the sharpest tool for that same habit, and it can be trained into everyone on the team.
There's one condition. You have to be willing to be the manager who writes a clear brief.
So next time the output is bad, don't open another model's tab yet. Reread your instructions, find which of the five is missing, and try again. If you want to practice this on real work with other people, join AI Circle.