AI Guides › Getting Started
By Nigel Guy · 2 min read
New users start by asking questions — one message, one answer, back to their own work to use it. That's a perfectly fine way to use it forever, but it's also only half the picture, and the people who get the most value tend to shift, at some point, from asking it things to delegating whole pieces of work to it. Those are genuinely different modes, and conflating them is why some people feel stuck at "it's a nice search box" long after there's more available to them.
The rule: asking gets you an answer to bring back into your own work; delegating hands over the work itself, including the parts you didn't explicitly specify — and it needs a different kind of instruction to go well.
| Asking | Delegating |
|---|---|
| "What's a good structure for this email?" | "Write this email for me, given these three points and this tone" |
| You do the actual assembling afterwards | It does the assembling; you review and adjust |
| A short, specific question is enough | You need to state constraints, not just the goal — what must stay true, what "done" looks like |
| Low risk if the answer's rough — you're filtering it yourself already | Higher risk if you don't review the output as carefully as you'd review your own draft |
| Works well from message one | Works much better once you've built some calibration on smaller tasks first |
Don't delegate a task you couldn't check yourself if it came back wrong — that's the "when not to use AI at all" territory, not a delegation problem. Don't delegate something with vague constraints and then be surprised the result missed a requirement you never stated.