The email draft looks fine. That's the problem.
You asked the AI to write a summary of your project proposal, and it came back polished, coherent, and almost entirely useless. Not because it's wrong, exactly. Because it's smooth. It has the confident tone of something that's been proofread by someone who has never met your actual stakeholders, your sceptical CFO, your colleague who will read the third line and immediately want to poke a hole in it. The AI gave you a clean surface. What you needed was something stress-tested.
Here's the move almost no one makes: once you have a draft, ask the AI to argue against it.
Not "what are some improvements?" Not "can you make this stronger?" Ask it to write a short, sharp case for why this proposal is wrong. Why the logic doesn't hold. What the CFO is going to say before they've even finished reading. Ask it to be the difficult person in the room.
Think of it like a sparring partner. The first draft is you shadow-boxing — fluid, uncontested, a bit too comfortable. The critique is when someone actually throws a punch.
That metaphor matters because it reframes what you're doing. You're not asking the AI to be clever or creative. You're using it as a mechanical sparring partner, one that can throw combinations it doesn't even know it's throwing. It doesn't have a stake in your proposal surviving. That's exactly what makes it useful here.
The output you get won't be perfect. Sometimes it'll identify a flaw that isn't really a flaw, or miss the one your CFO would land on in forty seconds. A failed sparring round is still a sparring round. You learn where you drop your guard.
There's a behavioural wrinkle here worth naming. It's called the endowment effect — we overvalue things just because they're ours. You wrote the proposal, or prompted it into existence, or edited it twice, so it already feels more solid to you than it actually is. The AI draft becomes psychologically yours pretty quickly. And once something feels like yours, you stop seeing its weak points. You start defending it.
Asking the AI to argue against it is a way of deliberately breaking that attachment before it costs you. Before you send it to the CFO. Before you say it in a meeting and have to watch someone's face do that particular thing.
The complication — and I want to be honest about this — is that it takes a bit of practice to write the counter-prompt well. "Argue against this" can produce something quite toothless if you leave it vague. The version that works is specific: "You are a sceptical finance director who has seen ten proposals like this fail. Point out the three most likely objections, bluntly." That's the prompt that gets you something useful. The vague version gets you a list of mild improvements with a polite tone.
That's the real skill being trained. Not prompting in some abstract sense. The specific, transferable ability to write a brief that puts the AI in a particular role, with a particular mindset, pointed at a particular job. You can use this in any meeting prep, any document review, any time you've been too close to something for too long.
Once you've done it a few times, something changes in how you work. You start writing first drafts with a small, nagging awareness: I'm going to sparring-test this in five minutes. That awareness, by itself, makes the first draft sharper. The shadow-boxing gets better because you know the sparring is coming.
That's the payoff. Not a better AI output. A better thinking habit, with the AI as the equipment.
This is the kind of workout inside The AI Habit — a daily training programme built around exactly this sort of skill: concrete, transferable, and a bit uncomfortable in the way that anything genuinely useful tends to be. Fifteen minutes a day. The first 14 days are free, so you can find out whether it's actually worth your time before you commit.
If the sparring-partner idea landed for you today, there's a whole session that walks you through it properly — with the prompts that work, the ones that don't, and the exact moment when most people give up and go back to the first draft.
https://harrysharman.com/projects/ai-habit/