---
title: '[Neurain Dev Log] #6. How I Actually Use the Brain These Days'
category: ai
tags:
  - ai
  - second brain
published: true
date: 2026-07-21 07:50:00
series: 'Neurain Dev Log'
seriesOrder: 6
description: 'I asked for a deck and got back more than I asked for, because it had pulled old meetings and past decisions out of the brain. Records compound. Past the two-month mark, output quality is decided by what has accumulated rather than by the prompt.'
---

I asked it to "draft a deck from this."

The draft that came back did not contain only what I gave it. Points raised in a meeting weeks earlier, a decision made long ago, and figures I had written up at the time were all folded in. I had not asked for any of it. I had also forgotten all of it.

That is the result of a structure built to search past records first whenever a request arrives. That day I felt the difference between an AI that does what it is told and an AI that knows my context.

## TL;DR

> *1) The secret to good output is not the prompt. It is the past records stacked in the brain.\
> 2) Records compound. Two weeks is a loss, a month breaks even, past two months it pays.\
> 3) The daily routine is three words: `_status` in the morning, `_save` during the day, `_wrap` in the evening.\
> 4) Memory flatters. Records do not. That is what earns you the right to write a retrospective.*

## 1. The full story of the deck incident

Taking the process apart: when I request a deck, the system searches my documents before building anything. It is the pipeline from post #4. Meeting notes, past decisions, and reference figures related to the topic get pulled up, and the draft is written on top of that evidence. So the output automatically carries two or three months of my context.

What surprised me most in that draft was one slide. It reflected a point that had come up briefly in a meeting weeks earlier and then been buried. I had forgotten I even saved those meeting notes. The system remembered that day's `_save`. When the system remembers what I have forgotten, some days the output is more accurate than my memory.

It used to work the other way. To build a deck I had to dig up the old material myself and feed it to the AI. Half a day finding, half a day building. Now the finding is zero. Only after finding disappeared did I realize that half of building had actually been finding.

There is plenty written about how to write good prompts. But after two months, what decides output quality is not prompt craft. It is the volume and quality of the records behind it. The same one-line request produces generalities when the brain is empty and produces something of mine when the brain is full. A prompt is a skill. A record is an asset. Skills are quickly learned by anyone. Assets belong only to whoever accumulated them.

## 2. Records compound

There is a warning I always attach when recommending this system. **The first two weeks are a loss.**

The effort of saving starts on day one, and there are no records to pull yet. There will absolutely be a stretch where you have built the thing and wonder what is supposed to be good about it. I went through it too.

Roughly how it felt over time:

| Period | How it felt |
|---|---|
| Weeks 1 to 2 | Only saving, nothing to pull. Pure cost |
| Weeks 3 to 4 | Occasional "wait, I saved that." Break-even |
| Month 2 | One request drags the whole past context along with it. Profit |
| After that | You cannot go back to working without it |

In investing terms it is a compounding curve. Boring at first, and past a certain point the slope stands up sharply. Which is why I think the hardest part of this system is not technical. It is surviving the first two weeks.

What got me through was designing saving to be easy. When saving, you do not classify, you do not agonize over a title, you do not choose a folder. You just throw it in. The AI organizes later. Once the psychological cost of saving approached zero, the deliberation disappeared entirely. **Records only started really accumulating from the moment I stopped asking myself "is this worth saving?"**

## 3. The daily routine

Morning. I sit down and type `_status`. A briefing appears, roughly like this.

> Unfinished: 1 report draft (resuming from where you stopped yesterday)\
> Ready to organize: 2 meeting notes saved yesterday\
> Status: search healthy, yesterday closed

The handover left by last night's me is read out by this morning's AI. The warm-up time spent recalling where I left off is gone. That warm-up used to consume an entire cup of coffee.

During the day I just work by talking. Drop in meeting notes, have material summarized, produce documents. When something comes up that must not be lost, one `_save`. It goes into the inbox exactly as it is.

Evening. `_wrap` closes the day. Today's accumulation is organized into knowledge, the handover for tomorrow's me is written, and a receipt appears confirming that what was saved actually turns up in search. Tomorrow morning's `_status` reads tonight's `_wrap`.

The unexpected effect of this routine was psychological. Work got an end-of-day. I used to leave the office carrying unresolved items in my head, repeating them so I would not forget. Now, the moment `_wrap` writes the unresolved items down, they leave my head. I handed them to tomorrow's me, so today's me is finished. I did not expect the first benefit of a second brain to be rest for the first one.

While this cycle runs, one day connects to the next without a break. The assistant that lost its memory every morning in post #1 now picks up yesterday every morning.

## 4. Nine lives running on one brain

A few of the nine areas, opened up.

**Work.** Strategy documents, meeting notes, and decisions accumulate in one area. When new notes come in, the related team documents get updated, and weeks later "why did we decide it that way?" comes back with an answer and its evidence. The deck incident happened in this room. The denser the work in an area, the more the brain is worth there, because a lot of context means a lot to forget.

**Studying the redevelopment.** The [guide to apartment floor-area terms](/en/2026/07/apartment-area-terms) on this blog is actually the first post to come out of this brain. Months of study were already organized in the area, so an article takes half a day. The experience of the distance shrinking between accumulating knowledge and producing something from it is more enjoyable than expected. What I studied becomes an article, and the article goes back into the brain as the basis for the next round of study.

**Investing.** I can pull up the reasoning behind a past judgment. Human memory keeps the conclusion and lets the reasoning evaporate. The brain keeps the reasoning with a date attached. It becomes possible to argue with the version of me from a few months ago. It is not that I was right then and wrong now. I can see what premises I judged on, so I only have to check whether the premises changed. The scariest thing in investing is not being wrong, it is not knowing why you were wrong, and a record at least spares you the second.

**Family.** I will only note that it exists. A private room is good precisely because it is private. To add one thing, the more sensitive the area, the more this system proves itself. Documents I would hesitate to upload to a cloud service sit on my own machine rather than anyone's server, and I have confidence that search only runs inside that room. That is what finally let me hand documents like these to an AI at all.

**The blog.** When an idea for a post occurs to me, I throw it in with `_save`. Half a sentence that came to mind on the subway, a link to an article, a question someone asked. Then on the day I decide to write, the brain lays out the accumulated fragments and related knowledge on one table. Writing from a blank page and writing with the ingredients already laid out are completely different jobs. Working this room taught me that on days when the writing will not come, the problem is usually the ingredients rather than the prose.

## 5. What still bothers me

Only saying good things turns this into an advertisement, so here is the friction.

**I cannot use it away from the desk.** The brain lives inside my MacBook. My phone cannot reach it. The convenience I had in post #1, running OpenClaw from Telegram while out, is something I have actually lost. Thoughts that come to me in transit end up in a notes app to be moved over later.

**There is only one computer.** Seeing the same brain from another device means solving synchronization, and I have not solved it. For now I get by on backups alone.

**It is hard to explain to people.** "Put everything in a folder and let the AI organize it" lands fine. After that, faces cloud over. Even with a one-line install, understanding the concept still has a barrier.

Those three come back as an improvement list in the next post. The honest state right now is that this is a brain that runs very well at a desk and is still half a brain away from it.

## 6. Swapping models in practice

These days I move between Claude and Codex. Some days Claude organizes the meeting notes in the morning and Codex picks them up in the afternoon to build the report. It works because both read the same folder and follow the same notices.

In post #1 I described switching models as moving house with nothing. Now it is the opposite. When a new model appears I do not switch to it, I plug it in. If it does well it stays, if not I pull it back out. Because the brain is in my folder, changing models became swapping a part rather than relocating.

There is a side effect too. I can put the same question to two models and compare the answers. One catching what the other missed happens more often than you would think. Just as I did development that way in post #3, cross-verification came free for everyday questions as well. News of a new model turning from dread into anticipation is the biggest psychological change this system gave me.

## 7. I write from records, not memory

The place this brain changed my work the most is writing. More precisely, **I stopped relying on memory**.

For example, this series contains a sentence about taking eight screenshots one night in late May. There is a passage about the scorecard falling from 0.941 to 0.890. I do not have any of that memorized. When I am writing and wonder when something happened, I ask the brain and it produces that day's record.

Why this matters: when a person writes about something from two months ago, they will inevitably flatter it. Writing this series, my memory and the record diverged several times. I remembered admitting one failure within a day and moving on. Opening the record showed I had clung to it for several more days. Every such passage was corrected toward the record.

**Memory flatters. Records do not.** I think that difference is what earns you the right to write a retrospective.

After two months with a second brain, the conclusion is one line. The real benefit is not memory. It is continuity. Yesterday's me can now hand work to today's me. That is all of it, and it is enough.