The best use of AI isn't getting an answer. It's getting unstuck.

There's a moment every designer and engineer knows well: staring at a problem that has no clean solution, where every option trades one kind of pain for another. Onboarding flows with too many things users must do. Dashboards with too much data and no obvious hierarchy. A checkout redesign where fixing conversion breaks trust, and fixing trust tanks conversion.

The instinct with AI tools is to ask them to solve it. Hand over the problem, get a solution back. That's the least useful way to use them.

Complex problems don't have answers. They have angles.

A genuinely complex problem — the kind worth calling complex — usually isn't missing a clever solution. It's missing a clear enough framing. The hard part is rarely "what's the answer," it's "what are we actually even optimizing for, and what are we willing to give up."

This is where AI earns its keep, not as an oracle but as a pressure-test. Describe the problem out loud (or in a prompt) and ask it to list the assumptions baked into your framing. Ask it to argue the opposite of your instinct as convincingly as it can. Ask it what a completely different industry would do with this exact constraint. None of these require the model to be right. They require it to be a fast, tireless sparring partner that doesn't get tired of being told "no, but what about—" fifteen times in a row.

A practical way to use it

1. State the problem without your preferred solution attached. Most people skip this and describe their answer instead of their problem, which locks the conversation into confirming a decision rather than examining one.

2. Ask for the strongest version of the option you don't like. It's easy to build a straw man of the choice you already rejected. A model has no ego invested in defending it properly.

3. Ask what's being traded away, not just what's being gained. Every real solution to a complex problem costs something. Naming the cost out loud is often the actual insight.

4. Use it to generate the boring alternatives too, not just the interesting ones. Sometimes the unglamorous option is correct, and it's worth having it on the table before falling for the exciting one.

What this isn't

This isn't "let the AI decide." Complex problems, almost by definition, involve values, tradeoffs, and context a model doesn't have — what your team can actually execute this quarter, what your users have been burned by before, what the CEO will and won't accept. Handing over the decision just moves the hard part somewhere it's less visible.

What changes is the shape of the thinking process. Instead of sitting alone with a hard problem until an idea arrives, you get a conversation — one where a tireless partner keeps handing you angles you hadn't considered, so the human sitting there making the actual call gets to make it with a fuller picture.

The goal was never to think less. It's to think from more directions before deciding.