AI problem solving
"What's the answer" and "help me think through this" produce different quality results — because they use the model completely differently.
Thinking partner, not answer machine
Ask "what should I do about this client situation" and you get a plausible-sounding recommendation, arrived at in a single leap. Ask "walk me through the factors that matter here, one at a time, before we get to a recommendation" and you get something structurally different — a reasoning process you can actually inspect, disagree with at a specific step, and correct before it compounds into a wrong final answer. Same model, same question underneath, completely different quality of output, because the second approach uses AI as a partner working alongside your reasoning instead of a machine dispensing a verdict.
The core technique: externalize the thinking, don't just request the output
Before asking for a solution, ask for the problem to be restated, the relevant factors to be listed, and the trade-offs to be laid out — then look at that intermediate reasoning before letting it proceed to a recommendation. Catching a wrong assumption at that stage is far cheaper than catching it after a full plan has been built on top of it.
Four problem types, four different approaches
Business problem
Lay out the constraints and stakeholders first, then generate options — not a single answer, several worth comparing.
A bug
Describe the expected versus actual behavior precisely, then work backward through what changed, rather than asking for a generic fix.
A math problem
Ask for the method to be shown step by step — an AI walking through its work is far easier to sanity-check than a bare final number.
A complex scenario
Ask what's actually uncertain before asking what to do — clarifying the unknowns changes what a good answer even looks like.
A structured loop worth defaulting to
Where this genuinely helps less
Problems that depend on tacit, hands-on experience — reading a room, a physical skill, a judgment call built on years of specific context nobody wrote down — don't hand off cleanly to a reasoning partner that's never been in the room. AI can still help structure the surrounding decision, but the core judgment stays yours.
The mistake that undoes all of this
Asking for the answer, getting it, and stopping there — skipping the intermediate reasoning entirely defeats the entire point of using AI as a thinking partner instead of a lookup tool. If you're not at least occasionally pushing back on a step in the reasoning, you're probably using it as an answer machine again without noticing the switch.
The short version
The single highest-leverage habit here is resisting the urge to ask for the answer first. Ask for the reasoning, inspect it, push back where something looks off, and let the final answer be the last step instead of the only one you see. That shift — from verdict to visible process — is most of what separates using AI well on a hard problem from using it carelessly on an easy-looking one.