The Dream-Build Loop: Capability Starts With a 45-Minute Walk
The distance between having an idea and having a working version of it used to be a funding cycle. For a lot of work, it's now about forty-five minutes. Most people haven't caught up to that yet — including people who use AI every day.
Here's an experiment I've started recommending to almost everyone, because the ratio of effort to insight is absurd.
Take a walk. Bring your phone. Pick one thing that annoys you about your own work — not a strategic priority, an annoyance. The report you rebuild every month. The email you write forty variations of. The list you keep re-sorting by hand.
Then spend the walk talking it through with AI. Not asking it to do the task. Describing what you wish existed, and letting it push back.
By the time you get back, you will very often have something. Not a polished tool — a draft, a structure, a first working version of an idea that was vapor when you left. I call this the Dream-Build loop, and once you've closed it a few times it's genuinely hard to stop.
Why the walk matters
The walking part sounds like a gimmick. It isn't, and I've thought about why.
Sitting at a keyboard puts you in execution posture. You start specifying immediately, and specification narrows you before you've explored anything. You end up building a slightly faster version of what already exists.
Talking while walking puts you in a different register. You describe the problem instead of the solution. You say "the annoying part is really that nobody ever knows which version is current" — which is a much better problem statement than the one you would have typed, and it's the kind of sentence that produces a genuinely different answer.
Try it
Start small. See what you can dream-build with AI in a forty-five minute walk. No ideas are too small — small is the entire point.
The loop is the mechanism
What makes this work isn't the AI. It's the loop closing. Dream, build, look at the thing, adjust the dream, build again.
Most people in large organizations have never experienced that loop at work. Their ideas go into an intake process. They get prioritized against other ideas. Six months later something arrives that resembles what they asked for, and by then the need has shifted. The loop is so slow that people stop having the ideas — not consciously, they just stop noticing the annoyances, because noticing has no payoff.
Close that loop to forty-five minutes and something changes in how people see their own work. They start noticing again. That's the shift I care about far more than any individual thing somebody builds.
Why "no idea is too small" is the whole rule
Every time I've watched this fail, it's failed the same way: someone picks something too big. They choose the problem they think is worthy of AI — the multi-team, multi-system thing that would be genuinely valuable to solve. Then they spend the walk discovering that it's hard, and conclude that this doesn't really work yet.
Small ideas close the loop. Closing the loop is what teaches you the shape of what's possible, and that intuition is the thing you actually needed. The big problems get easier later, once you know what the tool is genuinely good at — and, just as importantly, where it falls down.
There's also a confidence dimension I didn't anticipate. People who have built one small working thing behave differently from people who have attended training about building things. They ask better questions. They're more skeptical in a useful way. They stop treating AI as either magic or threat and start treating it as a tool with a grain, like every other tool they've learned.
From personal habit to organizational capability
The reason I care about this beyond individual productivity: it's the cheapest way I know to move an organization from AI fluency to AI agency.
Fluency is knowing how to use the thing. Agency is knowing you're allowed to change your own work with it — and having enough intuition to know what's worth changing.
You cannot train agency into people. You can only give them enough closed loops that they develop it themselves. Forty-five minutes at a time turns out to be a reasonable unit.
If you try this, I'd like to hear what you built — especially if it was small. Post it, or send it to me on LinkedIn. No ideas are too small.