Most people use AI the same way: summarise this, polish that, rewrite this email I've already rewritten three times. It's useful. It's also completely predictable, and it won't make your work stand out.
There's a more interesting approach that I came across recently, and it's changed how I use AI for anything important: stop using AI for answers. Use it to attack your work.
The angry CFO method
The idea came from a story about a senior consultant at Accenture who told a colleague to treat AI like a furious CFO with a personal vendetta against his presentation.
Not a helpful reviewer. Not a polite assistant. An adversary who thinks your proposal is fundamentally flawed and wants to prove it.
So the colleague tried it. He'd prepared a client deck and instead of asking AI to tidy it up, he prompted it to act as the CFO who thinks the whole thing is a disaster.
Three devastating objections came back. The kind that make you stare at your screen and think: fair point. He spent the evening fixing every weakness.
Next day, the actual CFO had no objections. Because every hole had already been patched, privately, before the real scrutiny arrived.
Why this matters more than "better prompts"
The AI conversation right now is dominated by prompt engineering: how to ask AI better questions to get better outputs. That's fine, but it misses something more valuable.
The most useful thing AI can do for your work isn't generating content. It's stress-testing the content you've already created. Finding the weak assumptions, the logical gaps, the objections you haven't considered. The things that would embarrass you in a meeting but that you can fix quietly at your desk.
Most people never do this because they use AI as a helper, and helpers tell you what you want to hear. An adversary tells you what you need to hear.
Helpers tell you what you want to hear. An adversary tells you what you need to hear.
How to use this in practice
Take any piece of work that matters, a proposal, a strategy document, a presentation, a business case, and feed it to AI with a prompt that gives it permission to be brutal. Some examples:
"Act as a sceptical board member. What assumption have I made here purely out of laziness?"
"Pretend you're a rival consultant who wants to steal my biggest client. What's the weakest part of this proposal?"
"Review this as the senior executive who's been told to cut 20% of spending. Why would you reject this without hesitation?"
"Be the grumpy end user who hates change. What part of this new system would frustrate you most?"
The prompts work because they give AI a specific, adversarial perspective. Instead of the generic "any suggestions for improvement?" (which produces generic, hedge-everything responses), you get targeted criticism from a defined point of view.
If your first reaction to the output is discomfort, that's the signal it's working.
The real skill this builds
Adversarial prompting isn't just a technique for improving individual pieces of work. It builds a habit of anticipating objections before they arrive. Over time, you start writing with those objections in mind. Your proposals get tighter. Your presentations preempt the hard questions. Your thinking becomes more rigorous because you've trained yourself to look for weaknesses rather than hoping nobody notices them.
The people who will get the most value from AI in the long run aren't the ones who use it to produce work faster. They're the ones who use it to produce work that's harder to argue with.
Try it this week
Pick one thing you're working on that genuinely matters. Feed it to AI and ask it to tear it apart from a specific adversarial perspective. Read the criticism. Fix the weaknesses. Then notice how much more confident you feel about the work.
Get criticised early and privately, rather than late and publicly. That's the principle. AI just makes it trivially easy to do.