EN / 中
← Writing
30 September 2026

The Brain Is an Orchestra

How I Built a Second Nervous System
Sharing the system at an AI hackathon pop-up in London
Sharing the system at an AI hackathon pop-up in London.

Most of us think of AI as a way to get more done. I built mine to protect the conditions for thinking.

Last Saturday I shared it at an AI hackathon pop-up in London. Thank you to She Nicest and Women in Science (WiNS) for the invitation.

The brain is an orchestra. It thinks in parallel, but I can only act on one line at a time. Speech, writing, work: one word, one task at a time. So the scarce resource is not tools, and usually not time. It is energy.

Every system I build starts there: guard the energy first, and output follows. Because I was running two ventures, twenty lessons a week, training, money, and content all through that one channel.

So I grew a second nervous system.

It works in parallel while I work in sequence. Different parts of the system record, remember, check, plan, and execute in the background. I keep only what actually needs me.

Five Voices, One System

My system has five parts: Apps, Knowledge, Judgment, Agents and Review.

The five sections of the system: Apps, Knowledge, Judgment, Agents and Review
Five sections, one score.

Apps open the system. They are its senses, recording what is actually happening in my life. There are three: Health, Teaching and Finance. Health gathers sleep, training, diet and heart rate in one place. Teaching holds lessons, student information, package balances and monthly invoices. Finance splits my spending across six accounts, so an overspend shows at a glance. The aim is not to collect more data. It is to make certain parts of my life visible enough that the system can respond to what is happening.

The three apps: Health, Teaching and Finance
The sensing layer: Health, Teaching and Finance.

My running data is a good example. The health app connects to my Strava data, Apple Health data, and food logs, then uses them to create training plans and give me advice based on how my body is doing right now.

The health app showing training and recovery
The health app reads my body before it gives advice.

Knowledge is its memory. I use Obsidian for what I know and believe, and Notion for what I am running and planning. Obsidian holds the 22 books that have shaped me most, 172 interlinked core notes and 607 connections between them. Notion runs more than 30 students, two projects, weekly plans and a handful of working databases. Before an important decision, AI can read both and bring the relevant pieces back to me.

Judgment is the part I refuse to automate. For expensive or irreversible choices, I keep a Board and an AI Court. The Board brings different viewpoints to the table. The Court challenges assumptions, tests the evidence and looks for holes in the reasoning. For smaller calls that need no debate, I have written twelve explicit rules, so I do not have to reason through the same problem every time it returns.

An AI Court session in July 2026 also showed me the most important layer of the system. Three sceptics questioned, from different angles, whether my market execution was sustainable. Five witnesses provided information on cash flow, health, visibility, execution and support.

The conclusion was simple: the real bottleneck is not money, and not time, but the remaining capacity of my nervous system.

That sentence changed how I understand the whole system.

Agents are the hands. Fifteen of them: one runs continuously, around twenty run on a schedule, and a dozen advisors I can call when needed. They handle my calendar, shopping lists, financial checks, information sorting and the many small, repeating tasks.

My goal is not to automate everything in my life. It is to reduce the small decisions that keep draining my attention. Agents do the work. I make the decisions.

Review is where the loop closes. The work of the fifteen agents feeds a weekly review and a monthly deep clean. The weekly review looks back at what actually happened, what changed, what needs attention and what comes next. The monthly clean goes deeper, checking the whole system and resetting the plan for the next stretch.

So the whole system forms a loop: the system acts, the system records, the system reviews, and then the system changes.

To me, this loop matters more than automation itself.

The results are visible and checkable.

Muscle mass, 33.4 kg last April to 47.2 kg this July.
Long runs, 15 km in May to 26 km this month.
Fastest pace over 5 km, from 5:25 down to 5:16.
Resting heart rate, from 63 to 53 in six months.

The System Is Also a Way of Thinking

The most interesting part of building this system is that it changed my relationship with AI. My creative process usually starts with handwriting. I use handwriting to dig an idea deeper. Then I speak to bring out the full structure quickly and go wider. Finally, I let AI organise it into modules, rules and workflows.

Talking about the system in an interview
Interviewed by coAI Club.

When the system makes a mistake, I do not just fix that one error. I try to turn it into a rule and write it back into the system, so the same problem does not have to be solved twice.

The whole process is: Handwrite → Speak → Let AI structure → Mistakes become rules.

So technology is only part of it. More importantly, I am building a new way for my thinking and the system around it to work together.

The Part I Care About Most: The Conductor

We often think of the brain as a whole, as if it completes one unified task. But much of our cognitive activity comes from the dynamic cooperation between different large-scale brain networks.

The three large-scale brain networks
Three networks, and the SN decides which one plays.

Very roughly, the Central Executive Network (CEN) is mainly involved in tasks that need focused attention, working memory and goal-directed effort. The Default Mode Network (DMN) is more involved in memory, imagination and mind-wandering. And the Salience Network (SN) helps us identify what matters most in the moment and plays a role in switching between cognitive networks and redistributing attention.

I think of the SN as a conductor.

The conductor plays no instrument. It decides who should be heard now.

This idea has an interesting parallel in the system I am building outside my brain. The problem is not that I need to keep myself focused all the time. Focus is only one state. Sometimes I need the CEN: writing, calculating, teaching, making decisions. Sometimes I need the DMN: wandering, making associations, connecting things that seem unrelated, imagining things that do not exist yet. And sometimes what I need to do is switch.

This is also what I think many modern productivity systems overlook. We are good at training focus and talking about deep work and efficiency. But constantly trying to stay in that mode can easily lead to burnout, so more and more people are recognising the importance of rest.

What interests me more is the transition between the two: when to do deep work, when to let the brain wander freely, and when I should stop making decisions because I no longer have enough capacity.

My second nervous system is really an attempt to externalise part of that switching.

To me, this is what a “second nervous system” really means. It is not meant to keep me running all the time. It helps me protect the energy that lets me think. As AI takes over more and more of the work of remembering, organising, searching and executing, what I want is not information, not tools, and not even just time, but the capacity to think.

大多数人把 AI 当成做更多事的工具。我造它,是为了守住思考的条件。

上周六我在伦敦一场 AI 黑客松快闪活动上分享了它。谢谢 She Nicest 和 Women in Science (WiNS) 的邀请。

大脑是一支交响乐团。它并行地思考,但我一次只能做一件事。说话、写作、工作:一个词,一个任务。所以稀缺的不是工具,通常也不是时间,而是能量。

我构建的每一套系统都从这里出发:先守住能量,产出自然会来。因为我把两个事业、每周二十节课、训练、钱和内容,全挤在同一条通道里。

于是我长出了一套第二神经系统。

它并行地运作,而我串行地工作。系统的不同部分在后台记录、记忆、检查、规划、执行。我只留下真正需要我的那部分。

五个声音,一套系统

我的系统有五个部分:应用、知识、判断、智能体、复盘。

系统的五个部分:应用、知识、判断、智能体、复盘
五个声部,一张总谱。

应用是系统的入口。它们是系统的感官,记录我生活里真正发生的事。一共三个:健康、教学、财务。健康把睡眠、训练、饮食和心率收在一处。教学管课程、学生信息、课时余额和每月账单。财务把我的开销分到六个账户,一旦超支一眼就看得到。目的不是收集更多数据,而是让我生活的某些部分足够可见,让系统能对正在发生的事做出回应。

三个应用:健康、教学、财务
感知层:健康、教学、财务。

我的跑步数据就是个好例子。健康应用连着我的 Strava 数据、Apple Health 数据和饮食记录,再用它们生成训练计划,并根据我身体当下的状态给我建议。

健康应用显示训练与恢复
健康应用先读懂我的身体,再给建议。

知识是系统的记忆。我用 Obsidian 存我所知道和相信的,用 Notion 管我正在做和规划的。Obsidian 里有对我影响最深的 22 本书、172 条互相链接的核心笔记,以及它们之间的 607 条连接。Notion 里跑着 30 多个学生、两个项目、每周计划和几个在用的数据库。重要决定之前,AI 可以读这两边,把相关的部分带回给我。

判断是我拒绝自动化的部分。遇到昂贵或不可逆的选择,我留着一个董事会和一个 AI 法庭。董事会把不同的视角摆上桌。法庭质疑假设、检验证据、找推理里的漏洞。至于那些不需要辩论的小决定,我写下了十二条明确的规则,这样同一个问题每次回来,我不必再从头想一遍。

2026 年 7 月的一次 AI 法庭,也让我看到了系统里最重要的一层。三位质疑者从不同角度追问:我的市场打法是否可持续。五位证人提供了现金流、健康、曝光、执行和支持方面的信息。

结论很简单:真正的瓶颈不是钱,也不是时间,而是我神经系统剩下的容量。

这句话改变了我对整套系统的理解。

智能体是系统的手。一共十五个:一个持续运行,约二十个按时间表运行,还有十来个顾问,需要时我随时能叫。它们打理我的日历、购物清单、财务检查、信息整理,以及许多小而重复的任务。

我的目标不是把生活里的一切都自动化,而是减少那些不断消耗我注意力的小决定。智能体做事,我做决定。

复盘是闭环合上的地方。十五个智能体的工作汇进每周复盘和每月深度整理。每周复盘回看真正发生了什么、什么变了、什么需要注意、接下来是什么。每月整理走得更深,检查整套系统,为下一段路重设计划。

于是整套系统形成一个闭环:系统行动,系统记录,系统复盘,然后系统改变。

对我来说,这个闭环比自动化本身更重要。

结果看得见,也查得到。

肌肉量,从去年四月的 33.4 公斤到今年七月的 47.2 公斤。
长跑,从五月的 15 公里到这个月的 26 公里。
5 公里最快配速,从 5:25 降到 5:16。
静息心率,半年内从 63 降到 53。

这套系统也是一种思考方式

构建这套系统最有意思的地方,是它改变了我和 AI 的关系。我的创作通常从手写开始。我用手写把一个想法挖得更深。然后用说的,把完整的结构快速带出来、铺得更广。最后,我让 AI 把它整理成模块、规则和工作流。

在采访里聊这套系统
coAI Club 的采访。

系统出错时,我不只是修好那一个错。我试着把它变成一条规则,写回系统里,这样同一个问题不必解决两次。

整个过程是:手写 → 说出来 → 让 AI 结构化 → 错误变成规则。

所以技术只是其中一部分。更重要的是,我在为自己的思考和它周围的系统,构建一种新的协作方式。

我最在意的部分:指挥

我们常把大脑当成一个整体,好像它在完成一个统一的任务。但我们很多的认知活动,来自不同大尺度脑网络之间的动态协作。

三个大尺度脑网络
三个网络,由 SN 决定此刻由谁来演奏。

非常粗略地说,中央执行网络(CEN)主要参与需要专注、工作记忆和目标导向的任务。默认模式网络(DMN)更多参与记忆、想象和走神。而突显网络(SN)帮我们识别此刻最重要的是什么,并在不同认知网络之间切换、重新分配注意力。

我把 SN 想成一位指挥。

指挥不演奏任何乐器。它决定此刻该听见谁。

这个想法,和我在大脑之外构建的系统有一个有趣的呼应。问题不在于我要一直保持专注。专注只是一种状态。有时我需要 CEN:写作、计算、教学、做决定。有时我需要 DMN:走神、产生联想、把看似无关的事连起来、想象还不存在的东西。而有时,我要做的是切换。

这也是我认为很多现代生产力系统忽略的地方。我们擅长训练专注,擅长谈深度工作和效率。但一直想留在那个模式里,很容易导致倦怠,所以越来越多人开始认识到休息的重要。

更让我感兴趣的是两者之间的过渡:什么时候做深度工作,什么时候让大脑自由走神,以及什么时候我该停止做决定,因为我已经没有足够的容量了。

我的第二神经系统,其实是一次尝试:把那种切换的一部分外化出来。

对我来说,这才是「第二神经系统」真正的意思。它不是为了让我一直运转。它帮我守住那份让我能思考的能量。当 AI 接手越来越多记忆、整理、搜索和执行的工作,我想要的不是信息,不是工具,甚至不只是时间,而是思考的能力。