Every AI training session I've sat in eventually hits the same moment: someone asks how to "write better prompts" as if it's a syntax to memorize, like a formula in a spreadsheet. It isn't. Prompting as a skill is delegation with extra steps, and the people who are already good at briefing a new hire are, without knowing it, already good at this.
I've watched non-technical managers outperform engineers at getting useful output from AI tools, and it took me a while to understand why. The engineers were trying to find the "correct" prompt, like there's a command that unlocks the right answer. The managers were doing what they do every day: explaining a task clearly enough that someone who doesn't share their full context can still produce the right result.
That's the whole skill. Prompting an AI model and briefing a subordinate are the same problem, specification, wearing two different outfits.
Both are specification problems
When you hand a task to a junior employee, you don't just say "handle the client email." You give them the account history, the tone the client expects, the decision they're not allowed to make without checking with you first, and what "done" looks like. Skip any of that and you get back something technically responsive but wrong in a way that costs you time to fix.
An AI model fails the exact same way, for the exact same reason: it wasn't given what it needed to do the job right, not because it's incapable of the job.
The parallel breaks down only at one point: a junior employee can ask a clarifying question if your brief is ambiguous. Most AI interactions, especially the one-shot kind, cannot. That makes the brief matter more, not less. Sloppy delegation to a human gets corrected in a follow-up conversation. Sloppy delegation to an AI model just produces a confidently wrong answer that looks finished.
The four things every good brief has
Whether the recipient is a new hire or an AI model, a brief that actually works has the same four ingredients.
- Context. What does the recipient need to know that they don't already know? For a new employee, that's company history, the client relationship, prior attempts. For an AI model, that's the same information, pasted in or referenced, because it has no memory of your business unless you give it one.
- Constraints. What are they not allowed to do? Budget ceilings, tone restrictions, things that must not be promised. Without constraints, both a junior employee and an AI model will fill the gap with something plausible-sounding that may be wrong.
- Examples. Show, don't just tell. "Write it like this" with an actual sample of the format you want saves more back-and-forth than any amount of describing the style in adjectives.
- Definition of done. How will you know the output is acceptable? A word count, a checklist, a comparison to a previous version. Vague requests get vague results, from people and from models alike.
Miss any one of these four and you get the same failure mode regardless of whether the "employee" is human or software: technically responsive, practically useless, and now you're the one fixing it.
Why this reframe matters for training
Most companies training staff on AI tools start with tool mechanics: which button, which model, which settings. That's backwards. The staff who will get real value from AI tools are the ones who already know how to brief well, and the fastest path to good AI use in a team isn't a tool tutorial, it's making delegation skill visible and teachable as its own thing.
I've seen this play out directly in projects, including automating repetitive back-office tasks where the AI component only worked once the team learned to write a request the way they'd write a brief for a new admin hire, not the way they'd type a search query. The technical setup took a day. Getting the team to specify clearly took three weeks of practice.
If you manage people, you already have the mental model. The gap most managers have with AI tools isn't a skills gap, it's a recognition gap: they don't realize the skill transfers.
A practical checklist to hand your team
Next time someone on your team says an AI tool "doesn't work" or "gives dumb answers," walk them through this before assuming the tool is broken:
| Delegation element | What to check |
|---|---|
| Context | Did you give the background a stranger to your business would need? |
| Constraints | Did you say what's off-limits, not just what's wanted? |
| Examples | Did you show the shape of a good answer, not just describe it? |
| Definition of done | Would you and the AI agree on what "finished" looks like? |
Nine times out of ten, the fix is in that table, not in switching tools or fiddling with settings.
The takeaway
Prompting as a skill is not a technical competency separate from how you already manage people, it's the same competency, applied to a collaborator that can't ask you a follow-up question. If you want your team producing better results from AI, don't send them to a tool tutorial. Send them back to the basics of a good brief: context, constraints, examples, definition of done. The people who already do that well with humans will do it well with AI on day one.