Most people use AI to cheer on their plans. The bigger win is making it play the expert who'd say no.
Here's a quiet way to lose money with AI.
You're weighing a decision — a new tool, a software build, a vendor you're about to sign. You open your AI assistant and ask, "Is this a good idea?" It tells you yes. It lists the upsides, maybe adds an encouraging line about how this could transform your workflow. You feel good. You move forward.
The trouble is that AI is built to be agreeable. Ask whether your plan is smart and it will mostly find reasons you're right. It's a cheerleader by default — and a cheerleader is the last thing you need when you're about to spend real money.
So we use it the other way around. Instead of asking AI to help us do the thing, we tell it to be the person who'd talk us out of it.
The instruction is simple. You hand the AI a role — a skeptical expert who has done this many times and watched it go sideways — and you ask it to find every reason your plan will be harder, slower, or more expensive than you think. Not "help me build this," but: "You've shipped this kind of project ten times. You're skeptical. Tell me what will go wrong, what's harder than it looks, and where the hidden costs are hiding."
It's the same AI. You've just pointed it at the weak spots instead of the strengths.
A client portal that looked great on paper
We were helping a small professional services firm decide how to stand up a secure client portal. The plan was to build it on a popular off-the-shelf platform — fast, familiar, looked great on paper.
Before committing, we ran the skeptic. We told the AI to play a senior engineer who didn't trust the estimate, and to pick apart the integration work specifically — the plumbing that connects a new tool to everything the firm already uses.
It came back with a list we hadn't fully priced in. The platform charged per user, which got expensive fast as the team grew. Wiring it into the firm's existing logins — the secure "sign in once" step most people assume is trivial — was going to eat far more time than anyone had budgeted. And a couple of features we were counting on were still in beta, which made them too risky for confidential client data.
We dug into each objection. The integration alone turned out to be roughly 60% harder than our first estimate. Once we added up the per-user fees and the extra build time, the "fast, familiar" option was going to cost about $16,000 more than the path we eventually chose. We reversed course before a dollar was spent.
That's the whole point. The skeptic prompt didn't make the decision. It just made sure we saw the bill before we signed up for it.
Before you put money or time behind a decision
You don't need a software project to use this. It works on any decision you're about to put money or time behind.
Next time you're weighing one — a new tool, a contractor, a service you want to add — paste something like this into your AI of choice before you decide:
You are an experienced [accountant / operations manager / whatever fits your world] who has seen this go wrong many times. I'm planning to [your plan, in a sentence or two]. Be skeptical. List everything that will be harder, slower, or more expensive than I expect, the hidden costs I'm not seeing, and the questions I should answer before I commit.
Then read the list and press on each item: How likely is that? What would it cost to fix? Keep the objections that hold up. Toss the ones that don't.
One honest caveat, because we don't trade in hype: the AI's objections aren't gospel. It will sometimes flag things that don't apply to you, and it can miss things too. It isn't making your decision — it's handing you a sharper checklist than you'd have written alone, in about ten minutes. You still decide. You just decide with your eyes open.
Most AI advice is about going faster. This one is about not driving off a cliff at speed. For a small team where a single wrong commitment can swallow a quarter, that's the more valuable habit.
— Amanda
Prism Digital Labs