Capture First, Think Later: How to Record Ideas So Your Agent Can Use Them

David Allen, who wrote Getting Things Done, put it in one line: “Your mind is for having ideas, not holding them.”

Most people agree with that and still work against it. When an idea arrives, they try to do two jobs at once: catch it and decide whether it’s good, where it belongs and what to do about it. The second job is slow and uncomfortable, so often the first one never happens. The idea gets judged, half-formed, in your head, and quietly dropped.

The fix is to split the work into two steps that happen at different times:

1. Capture, when the idea shows up: fast, messy, no judging.
2. Think, later, at a time you choose: sort, connect, decide, produce.

This used to mean a pocket notebook and a weekly review. Today the second step can be shared with an AI agent that reads everything you captured. That changes what good capture looks like, because now your notes have two readers: future you, and an agent that has none of the context in your head.

Here are seven habits that make captured ideas useful to both.

1. Speak, don’t type

The bar for capture has to be low enough that you clear it every time, while walking, between meetings, half asleep. Typing on a phone is slow, so you shorten the idea until it stops making sense. Speaking is roughly as fast as thinking, and you can talk your way through a half-formed thought until it takes shape.

Don’t worry about rambling. SpeakPen gives every recording a title and a summary and keeps the full transcript, so a messy minute of talk still becomes a note you can scan in a few seconds.

2. One idea per note

When you record “a few things I was thinking about,” you get one note covering three subjects with a title that fits none of them. Searching gets harder, and the agent pulls in noise every time it opens that note.

Record three short notes instead. They cost nothing, and each one gets its own title and summary.

3. Say why, not just what

“Usage-based pricing for teams” is a reminder you can’t do much with. Six weeks later neither you nor your agent knows why it seemed important.

Add the context you’ll have lost by then: what triggered the idea, what problem it solves, what you were looking at. A simple pattern that works:

What is the idea? Why might it matter? What is the smallest next step?

You don’t need to follow it word for word. Thirty seconds that cover those three questions beat a perfect sentence.

4. Use the names you’ll search for

Agents find notes by searching them. If you call it “the Q4 pricing project” in one note, “pricing stuff” in the next and “that thing Anna mentioned” in a third, they won’t connect. Say the project, client or product name out loud, the way you’d type it into a search box.

5. Record questions, not just answers

Some of your most valuable notes are questions: “Why do trial users drop off on day three?” “What would this look like if it had to work offline?” An open question is a perfect starting point for an agent. It can look through your other notes for anything related and come back with a first pass at an answer.

6. Think on a schedule, not when you capture

This is the second half of the split, and the half most people skip. Capturing without ever processing just moves the pile from your head to an app. That’s exactly how second brains turn into notes graveyards.

Put a fixed slot in your calendar, 20 to 30 minutes once a week, where you and your agent go through what you captured. The next post includes a weekly review prompt you can copy.

7. Let most ideas go, on purpose

Processing doesn’t mean acting on everything. Most ideas are worth less than they felt like in the shower, and that’s fine. The point of reviewing them is to decide: act now, keep for later, or drop. An idea you dropped on purpose doesn’t keep nagging at you the way one you just forgot does.

“Won’t an agent do my thinking for me?”

It’s a fair worry. Many people who lean on AI for everything say their own thinking starts to feel duller. The difference is in what you hand over.

Hand the agent the mechanical work: finding the notes, grouping them, spotting that you’ve said the same thing three times, laying out a first outline. Keep the judgment for yourself: which idea is actually good, what you believe, what to do next, what to drop.

A simple test: after a session with your agent, can you explain the decision in your own words, without looking? If yes, the agent helped you think. If not, it thought for you, so slow down and argue with its output before you accept it.

Your voice notes help here too. When the agent’s draft is built from your own spoken reasoning, you’re editing your own thinking, not adopting someone else’s.

Why this matters more with an agent

An agent can only work with what you gave it. A notebook full of three-word reminders gives it almost nothing. A few hundred spoken notes, each with one idea, its context and the names you use, give it something to reason about: patterns across months, connections you missed, drafts built from your own thinking instead of generic text.

Capture is the part only you can do. Make it fast and frequent, and let the thinking happen later, with help.

Previously: Five moments your best ideas get lost. Next: Turn voice notes into output with your AI agent.