Every real thing I've built started the same way: as a swirl. A half-formed idea that lived in my head, felt exciting at 11pm and impossible at 7am, and stayed stuck because "figure out the whole thing" is not a task anyone can start.

AI turns out to be exceptionally good at un-swirling. Not because it has the answers, because it asks you the questions in the right order and holds the pieces while you think. Here's the process.

Step 1: Brain-dump without structure

Open a chat and empty your head. Voice memo it if that's easier. Don't organize, that's the machine's job:

I'm going to dump a messy idea on you. Don't give me advice yet. Just listen, then reflect back: what is the core idea, what am I excited about, and what am I clearly afraid of? Here it is: [everything, unfiltered]

That reflection step matters. Seeing your own swirl organized into "the idea / the excitement / the fear" is the moment it stops being fog and starts being material.

Step 2: Let it interview you

Now interview me. Ask me one question at a time, the questions a sharp, kind advisor would ask before helping me plan this. Don't move on until I've answered.

One at a time is the trick. A list of twenty questions is homework; a conversation is momentum. Expect questions about who it's for, what it costs, what you'd need to learn, what done-in-six-months looks like. Answer honestly, including "I don't know", those become research tasks, not roadblocks.

Step 3: Demand the boring version

Ideas die of grandiosity more often than failure. So ask for modesty:

Based on everything I've told you, give me the smallest, most boring version of this that could exist in 30 days. Not the dream version, the version that tests whether the dream is real.

The boring version of a bakery is selling twelve loaves to neighbors. The boring version of a consultancy is one paid conversation. Small is not a compromise; small is evidence.

Step 4: Stress-test it

Now argue against this plan. What are the three most likely reasons it fails? Be specific and unsentimental. Then tell me which of those three I could cheaply test in advance.

You want the objections now, from a machine, at zero social cost, not in month four from reality. Notice which criticism stings most. That sting is information.

Step 5: Convert to a two-week runway

Turn this into a plan for just the next two weeks. Rules: no step longer than 45 minutes, each step has a specific output, and the first step is something I can do today.

Not a roadmap. A runway. Two weeks of 45-minute steps is something a full life can absorb, early mornings, lunch breaks, the pocket after bedtime.

Step 6: Keep the thread alive

Keep the whole conversation in one chat and return to it weekly: "Here's what I actually did and what I learned. Update the plan." The plan that adapts to your real weeks is the one that survives them.

You don't need more confidence to start. You need a smaller first step. Confidence is what you find on the way.

When the idea involves leaving something, a job, a career, pair this with the career-change planning guide.