Gotchas, Limits & Anti-Patterns

Gotchas, Limits & Anti-Patterns

When "Just Ask Cowork" Is the Wrong Answer

Not every problem is a prompting problem.

You've got a decision to make (which of two vendors to renew with) and you're tired of it. So you paste both proposals into Cowork and ask it to decide. The answer is reasonable, but it leans on numbers that aren't in the documents. You re-prompt with more context. Better, but it's hedging. You re-prompt again, sharper this time, demanding a single recommendation. You get one. It sounds confident. You still don't trust it, because you know it doesn't know what you know about the account manager who ghosted you in March.

You spend twenty minutes wordsmithing the prompt. The decision is no closer.

This was never a prompting problem. No phrasing was going to fix it, because the bottleneck wasn't Cowork's understanding. The call was yours to make.

Once a tool gets genuinely useful, every problem starts to look like a prompt. That reflex is mostly a gift. Sometimes it's a trap.

#The prompting reflex

When Cowork works, it works often enough that you stop checking whether a task belongs to it at all. A bad result reads as a bad prompt, so you reach for the obvious fix: prompt harder. More context, more structure, more re-rolls.