On Thinking Out Loud

The Smartest Room You've Never Walked Into

Why brainstorming is the most underused AI technique, and what you're actually tapping into when you do it.

May 2026 · 6 min read

Most people use AI the way they use Google. Type a question, get an answer, close the tab. That's fine. But it misses something far more interesting, and honestly far more valuable.

I've been using AI as a thinking partner for long enough now to notice a pattern. The sessions that produce the most value aren't the ones where I asked for information. They're the ones where I showed up with a half-baked idea and left with a clearer head. Not because the AI handed me a solution, but because the back-and-forth forced one out of me.

That's what this post is about. Brainstorming as a tool, and what happens when you start using it with AI deliberately instead of treating the whole thing like a search engine.


The technical bit (bear with me)

What's Actually Happening Under the Hood

Here's the boring explanation of how a large language model works: it predicts the next token. That's it. Given some input, the model calculates the most statistically likely continuation based on weights trained on an enormous amount of text. No magic. No understanding in any mystical sense. Just math, at scale.

Now here's the interesting explanation.

Those weights were trained on essentially everything humans have ever written. Books, research papers, forum debates, startup post-mortems, naming contests, market research reports, philosophical treatises, product reviews, and everything in between. When you open a chat window with an AI, you're not querying a database. You're sampling from a compression of collective human thought.

Think about who's in the room

Every naming consultant who ever wrote about brand strategy. Every market researcher who published their methodology. Every founder who blogged about the decision that nearly killed their company. Every philosopher who grappled with the same kind of question you're wrestling with right now, just in a different century and a different language.

That's not a chatbot. That's a brainstorming partner unlike anything that's existed before.

Both framings are true at the same time. The math is real. But so is the philosophy. And which one you hold in mind when you open that chat window shapes how you use it.


The part nobody talks about enough

The Asymmetry Is Historic

For most of human history, access to deep expertise required money, connections, or luck. If you wanted to think through a market entry, you hired a consultant. If you wanted sharp feedback on a product name, you commissioned a focus group. If you wanted to stress-test your strategy against someone who'd been there before, you needed to actually know someone who'd been there before.

Most people didn't have those resources. Most ideas died in someone's head because there was no one useful to think out loud with.

Now there is. And people are treating it like a slightly fancier search engine.

I've used it for market research, the kind that used to take days of reading and synthesizing before I even knew what questions to ask. For naming things: products, projects, internal tools, concepts that needed a handle before they could be communicated. In both cases, the value wasn't the AI generating an answer. It was the dialogue forcing clarity. You can't have a productive brainstorm about something you can't articulate. The process of articulating it is half the work.

The value isn't the output. It's what the process forces out of you.

The rubber duck, upgraded

Why It Works Even When the Answers Are Wrong

I've done this my whole career without knowing it had a name. Talking through a stuck problem with a coworker, not because they had the answer, but because saying it out loud helped me find it myself. Cornering a friend or family member at dinner with something I was wrestling with. Narrating a problem to no one in particular on a long drive. And yes, occasionally thinking out loud at my desk, which I used to chalk up to being weird and now consider entirely justified.

Turns out, this has a name. In software development it's called rubber duck debugging: you explain your problem out loud to an inanimate object (traditionally, a rubber duck on your desk), and the act of articulating it often reveals the solution. The duck doesn't respond. You just need to hear yourself talk.

AI is that, with one meaningful difference. The duck talks back. And it's read everything.

The most productive sessions I've had weren't the ones where I asked AI for solutions. They were the ones where I specifically asked it not to give me solutions yet. Before you suggest anything, ask me five questions you'd need answered to give me useful input. That single shift changes the dynamic. Instead of consuming output, you're using AI to excavate your own thinking.

And even when AI is wrong, it's often usefully wrong. A bad suggestion that makes you say no, that's completely off, and here's why is still doing real work. It's helping you discover what you actually believe.


Some things worth trying

Brainstorming With AI Is a Skill. A Few Approaches Worth Experimenting With.

Doing this badly (asking for a list of ideas and picking the first one) is probably worse than not doing it at all. You end up with generic outputs optimized for the average, not your specific situation. A few things that tend to help:

1
Ask it to ask you questions first

"I'm trying to decide X. Before you suggest anything, ask me five questions you'd need answered to give me useful input." It tends to surface things you didn't know were unclear, which is often the real problem anyway.

2
Change the vantage point with roles

"Act as a skeptical customer who's heard this pitch ten times before." "Act as someone who tried this two years ago and failed." Different personas pull different knowledge to the surface, and often remind you that the collective knowledge you're tapping includes people who disagreed with each other.

3
Steelman the opposite of your instinct

Whatever direction you're leaning, ask AI to make the best possible case against it. Not to be contrarian, but to find the holes before you commit. It’s a useful pressure test before you’ve invested too much in a direction.

4
Push back on generic

When you get an obvious answer, say so. "That's the first thing anyone would say. What's the non-obvious version?" The model has more range than it shows by default. It usually just needs a push.

The failure mode to avoid

Bad brainstorming with AI looks like: ask for a list, skim the list, pick something, move on. You've outsourced your thinking instead of sharpening it. The goal is a back-and-forth where you do the thinking and AI plays sparring partner, not a vending machine for ideas.


Making it a habit

The Five-Minute Rule

The shift I'd suggest isn't dramatic. It's almost embarrassingly simple: before you Google something, brainstorm it. Before you start a new project, spend ten minutes with AI talking through what you're actually trying to accomplish. Before you make a decision, run it past the collected wisdom of everyone who's ever thought about this kind of decision.

Think of it as a morning standup with yourself, but with a partner who's read everything and forgotten nothing. Five minutes. One problem. No agenda except thinking out loud.

Try this tomorrow

Pick one thing you're genuinely uncertain about: a decision, a direction, a problem you've been circling. Open a chat, say "I want to brainstorm something, don't give me answers yet, just ask me questions." See where it goes.

The sessions that feel unproductive in the moment are often the most useful. Not because you found the answer, but because you found the right question.


Everyone Who Ever Thought About This

The next time you open that chat window with a half-baked idea, remember what you're actually sitting across from. Not a product. Not a tool. A compression of human thought: philosophers and practitioners, researchers and founders, people who got it right and people who got it spectacularly wrong and wrote about it afterward.

For the first time in history, anyone can brainstorm with all of them at once. You don't need a budget for consultants, a network full of the right people, or an office with a whiteboard and colleagues who have time for your shower thoughts.

You just need to show up with a problem and be willing to think out loud.

That's worth more than most of us are treating it as.