Process your brain-dump into next actions

A captured inbox is a relief, not a system, until every item becomes a clear next action

You are at the Clarify step of the Empty-Head Loop. Capture got the chaos out of your head, but a full inbox is a pile, not a plan, and a pile you never process is just your head with extra steps. Learning how to process a brain dump is the move that turns raw capture into a system you can run a week from. The shift here is that you stop sorting by hand and stand up a triage agent to do it.

By the close you will have a repeatable processing routine, a working triage agent that tags and routes the dump, an organised second-brain workspace holding the output, and an empty inbox.

Processing means deciding, one item at a time, to zero

Processing is taking each inbox item and making exactly one decision about it, top to bottom, until the inbox is empty. You never skip an item and you never re-file it back into the inbox. For every item, ask one question: what is this, and what is the very next physical action?

Each item resolves into one of four outcomes:

• Next action, if it needs doing and you can name the single next step ("email Sarah the revised quote").
• Project, if it needs more than one action to finish ("launch the new pricing page"), which goes to your project list with its own next action.
• Someday, if it is real but not now ("learn Figma auto-layout"), parked where you will see it on review, not lost.
• Delete, if it is no longer relevant, which is far more common than people expect.

The two-minute rule sharpens this: if a next action takes under two minutes, do it now rather than tracking it, because the tracking costs more than the doing.

Stand up the triage agent: the actual setup

The sorting above is exactly the kind of repetitive judgement an agent does well, so you hand it over. Your second brain lives in Notion, a workspace with a captured-items database (the raw dump synced or pasted in) and four destination views: Actions, Projects, Someday, Archive. The triage agent reads each raw item and proposes the outcome, the rewritten next action, a project link and a context tag.

Give the agent a rules prompt it runs against every item. A worked version:

You are my triage agent. For each captured item, return JSON with: type (one of action, project, someday, delete), next_action (a concrete verb-first next physical step, or null), project (the parent project name if it belongs to one, else null), context (one of @computer, @calls, @errands, @deep), and est_minutes. Rewrite vague items into a clear next action. If an item takes under two minutes, type=action and flag do_now=true. If it is no longer relevant, type=delete. Never invent commitments I did not capture.

You run the dump through the agent, review its proposals in a single pass, correct the few it gets wrong, and accept the rest into the Notion views. What was an hour of manual sorting becomes ten minutes of supervising. The agent never decides anything irreversible; it proposes, you approve, the inbox empties.

Notion plus a triage agent: a worked example

Take a freelance management consultant running a one-person practice, the sort with a perpetually overflowing notes app and a Sunday-night ritual of dreading the week. The challenge was that processing the backlog by hand took about ninety minutes every weekend, so it kept getting skipped, and a skipped week meant the inbox calcified into a 200-item swamp.

The approach was a Notion captured-items database fed from the inbox, plus the triage agent above run over the backlog. The agent tagged and routed all 200 items in one batch; the consultant spent twelve minutes reviewing, correcting roughly fifteen percent of the calls, and accepting the rest.

The measurable result: weekly processing time dropped from ninety minutes to about fifteen, the inbox has hit zero every week since, and the consultant retired the Sunday dread because the swamp can no longer form.

INTERVIEW EWOUD: When you first handed sorting to a triage agent, what did it get wrong that you had to teach it, and where do you still insist on deciding by hand?

Pitfalls at the processing stage

Letting the agent auto-commit. An agent that files straight into your action list without your review will quietly invent or mis-scope tasks. Keep the propose-then-approve gate.

Filing without a next action. "Website redesign" in your action list is not actionable. Every action item names a verb-first next physical step.

Processing into the same inbox. Output goes to Actions, Projects, Someday or Archive, never back into the inbox you are trying to empty.

Skipping the delete. A processing pass that deletes nothing is timid. Most stale loops should die here.

With an empty inbox and every loop now a clear next action in your second brain, you have a clean list of real work. The next move is to give that work a time and a place, with a scheduling agent blocking your week.