Turn a Brainstorm Into a Working Spec

That's not a badly run brainstorm. That's what thinking with another person looks like, and it's why the idea is sharper at the end than it was at the start.

Nobody books time for this, either. It happens because a friend asked a good question at a café table. And you don't know it's one of those conversations until you're already ten minutes into it — that's the real problem, because by then getting set up means stopping.

Pastok starts from a complication on your Apple Watch face. The moment you notice this one's worth keeping, one tap and it runs from there, through the rest of it and the walk to the station afterwards. That walk is usually where the better version shows up. Opening a laptop mid-argument would have ended the argument. You're recording another person, so ask: "mind if I record this? Just for my own notes." Four seconds, and if the answer is no, it's no.

Pastok's Activity gives you the conversation with its structure visible. Roughly: here's how it opened and what the initial read was, here's where the pushback came and what it was about, here's the part you went deep on, here's what you landed on, here's what's still hanging. You can see the shape of the ninety minutes without replaying them.

Which means a few things you'd normally lose are just sitting there:

Where the idea moved. It started as one thing and ended as another. The path matters, because the reason you abandoned the first version is usually the strongest argument for the second.

What you settled versus what you merely said with conviction. These feel identical two days later.

What nobody resolved. The questions you couldn't answer between you. This is the agenda for the next conversation, and later on it's what stops a coding agent from quietly inventing answers on your behalf.

What you ruled out, and why. Ruling something out never feels like information at the time. It feels like the conversation moving on. Then next week you re-add the feature you'd already killed, because all you kept was the conclusion and not the reason.

The tangent that turned out to be the point. Sometimes that's the offhand thing at minute fifty. It didn't feel important yet, so you wouldn't have written it down.

The one that surprises people is prior thinking. You've almost certainly circled this idea before. Six weeks ago on a walk, maybe, or in a different conversation with someone else entirely. You've forgotten the specifics. But Pastok searches what was said by meaning, so you can just ask:

Have I talked about anything like this before? Pull up what I said and how it differs from today.

Now the idea isn't starting from today's conversation. It's an accumulation, and the version you had six weeks ago often contains the objection today's version hasn't dealt with. Ideas rarely arrive in one sitting. They get re-argued with different people, and your position moves between them:

Everything I've said about the pricing model across these conversations, and where my position changed.

Then you can push on it with all that context loaded:

What did we not consider? Where is this thin?

Take the open questions and lay out what each one depends on.

Chat can go and look things up, too, and that's worth pairing with the open questions specifically. Those are the ones neither of you knew the answer to. Whether someone's already built this. What the going rate is. Whether the API you both assumed exists actually does.

One of us thought there's already a tool doing this. Go find out, and tell me how it differs from what we described.

Asked in a blank chat window that's a generic question and you get a generic answer. Asked here it's aimed at your version of the problem, because the context isn't a paragraph you retyped from memory. It's the actual argument, including the parts you'd have summarised away.

If you do this often enough, write the format down once as a Skill so you're not rebuilding it each time:

From these conversations: the problem in a paragraph, who it's for, what we settled, what's explicitly not being built and why, and the open questions left as open. Don't resolve anything we didn't resolve.

A Skill runs when you invoke it, on a schedule, or when Pastok notices a situation. Output is Markdown or PDF.

One check before you rely on any of it. A model will flatten "we could maybe do it that way" into a requirement without changing tone, so confirm that anything recorded as decided was actually decided. Speaker separation gives you structure rather than certainty too — one of our users had a friend's problem written up as his own. The full transcript is on the web app at app.pastok.com when you want to check what was said.

• Open the Pastok web app

Paste that into Claude Code or Cursor and it's a better brief than most people write from a blank page. The "don't build this" list and the open questions especially. Those two go missing more than anything else, and they're the two that cost you the afternoon.

Pastok doesn't do that step automatically. It doesn't write code, doesn't touch your repo, doesn't push anything anywhere. What it does is make sure the ninety minutes where the idea got good are still there on Thursday, in a form you can build on.

• How to Turn Voice Notes Into a Searchable Personal Knowledge Base