Second Brain is over. Build the Brain.

Second Brain is over. Build the Brain.

1. Most work is not work.

I've been in startups since I was twenty. After that, I saw how big companies actually operate. E-commerce, software, consulting, advertising, manufacturing — I've watched "work" happen across a lot of industries. They all share something.

Most of what we call work is not work.

Writing technical docs carefully. Revising a BRD again and again. Cleaning up meeting notes. Writing follow-up emails. Posting status updates in Slack. Lining up the agenda before a meeting. Summarizing decisions after one. Leaving review comments, replying to those comments, and then having someone summarize the thread once it hits thirty replies.

It looks like work. It feels busy. By the end of the day, you're tired. But the actual value created has barely moved.

In the old org, this made sense. Get the design right and you cut rework. Reach alignment early and the team moves faster. If the foundation drifts, nothing built on top of it holds. So you write carefully. You check twice. That was right.

Was. Past tense.

Time spent typing documents by hand is no longer value. It's work that looks like work. The only real value is in judgment. What to build. Who to work with. When to walk away. That's the part humans should keep. Everything else is transferable.

So the question I keep coming back to is one thing.

How do you promote everyone out of the busywork?

This isn't about eliminating the busywork. It's about putting a new layer underneath it. A layer that absorbs it. Once that layer exists, every human moves up one level. Everyone becomes the person who directs, approves, decides. The org chart doesn't flatten. A new tier appears below it, and the humans get pushed up.

This is what AI should be doing. And it's mostly not happening yet.

2. Why "mostly not happening"

Since 2023, we've been told AI is going to change work. What's actually happening on the ground is mostly "drafting got faster." Email drafts. Doc drafts. Code drafts. Image drafts. Drafting did get faster, and that's not nothing. But has the number of tasks gone down? It hasn't. For most people, it's gone up.

Why?

Because today's AI tools have no origin. They start when you say "write this." Until then, they're silent. Finding the work that needs doing is still your job. What's needed, what's urgent, what's due, what to delegate — all of that is still you. That whole front end of the work is still human.

And that's the part that actually drains people.

Doing a hundred well-defined tasks, once you're warmed up, doesn't really tire you out. What tires you out is deciding every morning what should even be done. Reconstructing what happened yesterday. Predicting what's coming next week. That's cognitively heavy. As long as humans are the ones doing it, no one becomes the director.

So what's needed next is to move the origin of work to the AI side.

Not the human providing context to the AI. The AI side already has the context. The AI side surfaces "here's what you should do now." The human approves, rejects, or edits. That's the drafting workflow, and that's the moment the org actually moves up a level.

The catch is, for the AI to have context, it needs to be watching something.

3. The four layers of AI for knowledge work

Let me lay out the layers as I see them.

L0 — Search. Make what you wrote findable. Smart grep across your docs, your notes, your messages. Useful. The bar is low. This is becoming table stakes. Hard to differentiate on.

L1 — Recall. Record the day — screen, audio — and answer "what did we decide on Tuesday?" later. External memory. A lot of capital has gone here. Several startups have launched on this premise.

But this is fundamentally passive. It answers when asked. The number of tasks doesn't go down. Recall just gets easier.

A useful feature. But it doesn't make anyone a director. It just gives a department head a slightly better assistant.

L2 — Drafting. The system watches the day, learns the shape of your work, and starts writing the next document, the next reply, the next ticket on its own. The human becomes a reviewer, not an author.

This is the layer where the number of tasks finally drops. AI writes, human reads. AI proposes, human picks. AI follows up, human decides. The org moves up.

Almost no product has reached L2. Everyone stops at L1. The reason is that L1 ships cleanly: record, index, build a nice UI, you have a product. L2 requires something harder on top — turning observation into hypothesis, hypothesis into draft, draft into something a human can approve in seconds. That's a different kind of design problem.

L3 — and at L3, it stops being a tool.

It's an OS.

Why use the word OS. There's a reason.

4. Taking apart "Second Brain"

At some point, "Second Brain" became a phrase people used. The framing: a backup memory for your real brain. Take notes. Tag them. Review them. Organize. Cultivate.

It's a beautiful idea. I was one of the people who got pulled in for a while. I kept neat notes. I enjoyed designing tag systems.

But this idea rests on one assumption.

That the first brain — the biological one — is at the center of work.

The notes assist it. The tools are subordinate. The human is the protagonist. Digital plays a supporting role. Digital exists to patch the weaknesses of the real brain — forgetting, capacity, lack of search. That's the Second Brain worldview.

That assumption isn't true anymore.

Look at where knowledge work actually happens today. Calmly, just observe yourself. What did you do today? You thought through something in Slack. You shaped it in a design tool. You decided in a tracker. You documented it later. Between meetings, you replied to messages, you had twenty browser tabs open, you watched a chart on a second monitor, you sent three parallel tasks to three different agents.

Almost nothing is happening entirely in your head.

The brain is still there as an input device and as a final-judgment device. "Let me write this." That's the brain. "No, this direction is wrong." That's the brain. But the thinking itself is happening on the screen. You write while thinking. You design by moving things. You decide by looking at material.

The physical location of cognition has moved. It's no longer inside the skull. It's on the screen.

So the "first brain plus second brain" model is misaligned with reality.

The actual structure: the working brain is on the screen, primary. The biological brain is what handles input and final approval.

"Second Brain" is a kind word. It keeps the human as the protagonist. It says: you're the lead, digital is supporting cast. But that kindness is what's blocking the real leverage.

Because as long as the human is the protagonist, the human still has to do everything. Collect the information. Organize it. Recall it. Connect it. Decide on it. All of it, the human. Digital just makes that "easier." It can't actually do it for you.

The endpoint of Second Brain is a beautifully organized note system. But however well-organized it is, the one using it is you. The one pulling things out is you. The one acting on what's pulled out is you. The cognitive load doesn't drop. Sometimes it goes up, because now you also have to maintain the system.

I'm dropping that word.

Not the second. Rebuild the first. Redesign the brain itself.

That's what SHOGUN is.

5. What it actually means to build the brain

What does it mean, concretely, to build the brain. Three things.

5.1 Change how context is captured

First: change how context is captured, at the root.

Today's tools can only work with what the human explicitly wrote down. Notes, docs, files, saved chat logs. Search AI, recall AI — they're all dealing with "something already saved."

But eighty percent of decisions never become files.

The moment you opened five tabs and compared them. The moment you skimmed a thread and thought, "I'll come back to this." The moment you started replying, wrote three lines, deleted them. The moment another window pulled your attention, and when you came back you'd lost the thread. The design you opened in Figma for thirty seconds and rejected. The Slack message you drafted and never sent.

All of it disappears in place. The real context of work isn't in what was written down. It's in the stream of operations that happened.

The fact is already there. It's happening on the screen. The problem is just that nothing is capturing it.

There's a sloppy answer here, and a correct one.

The sloppy answer is: record the whole screen. Video plus audio. Technically, easy to do. In practice, it breaks. Storage breaks — a day of full-resolution video is many gigabytes. Privacy breaks — being recorded continuously makes users freeze up. Searchability breaks — pulling "that one moment" out of raw video is the same as watching it again.

The correct answer is to capture the meaning of the operation, not the pixels. Not what's on screen. What's being done. Which app, which document, how long, how it was edited. Who, about what, with what conclusion. Not pixels — a structured stream of meaning.

It's light. It's searchable. The privacy granularity is controllable. And agents can read it.

That becomes the input device for the new brain.

5.2 Treat agents as first-class citizens

Second: treat agents as first-class citizens.

Today's doc tools, note tools, task trackers — they're designed for humans only.

Clean UI. Readable type. Drag and drop. Animations. These have real value for humans. For an agent, they're noise. What an agent needs is structured data, a writable API, a change history, and explicit permissions.

When you bolt agents onto tools built for humans, both sides operate at half capacity. The agent has to scrape the UI. The human has to retrace what the agent did. Neither one is actually using its full capability.

The right design is one underlying data layer with two faces.

For agents: machine-readable raw context, explicit capability definitions, clear read/write boundaries. Hand it to them directly.

For humans: maximally compressed summary, highlights only on the parts where judgment is needed, a three-button UI of approve, reject, edit. Squeeze the information. Show only what matters.

Same data. Two faces. That's the OS's job. It's not "a human app with agent features bolted on." It's a system designed from the start for both.

5.3 Move the origin to the AI side

Third: move the origin of work to the AI side.

Touched on this earlier. Going one level deeper now.

Putting the origin on the AI side means: the AI side is the one saying "here's what's needed now."

Open your computer in the morning. The things you should do today are already laid out as proposals. Yesterday's continuation. Unanswered threads. Three decisions waiting. Two people you've kept waiting. One thing due tomorrow. Each one comes with a draft attached. You read. You approve, reject, or edit.

This isn't a to-do list. To-do lists are written by humans. This is different. The AI side proposes, based on observation.

You don't have to start your morning by reconstructing "what am I doing today" from scratch. The only thing left to think about is: from this stack of proposals, what to prioritize, what to drop, what to add. That's it.

That's what "promoting everyone out of the work" actually looks like in practice.

A CEO doesn't write their daily task list from zero. Their chief of staff brings proposals. The CFO brings proposals. The COO brings proposals. The CEO decides. SHOGUN puts every individual into that position.

6. Why the value compounds over time

And there's a structural reason this gets more valuable over time.

Ten years ago, work fit on one screen. That was enough. Office, mail, browser. Done.

Today, an AI-native individual runs two, three, four monitors. Traders watch charts and feeds at the same time. Engineers run code, preview, docs, terminal in parallel. Operators of agents juggle five conversation threads. Designers move between Figma, references, specs, feedback — and never quite save a file.

The next wave is already coming. Spatial computing. Once headsets are common, the very concept of "a screen" changes. Eight, ten surfaces become normal. You switch with your eyes, place with your hands, pin things in space.

Human attention does not scale with screen count.

This is a structural limit of the brain. Working memory is roughly the same whether you have one screen or ten. You can track maybe four threads of context, focus deeply on one. The cognitive cost of context-switching grows superlinearly with surface area.

The wider your surface, the harder it is to see the connections between actions. Which screen you were just on — gone. What that conversation was leading to — can't recall. Run three projects in parallel and one of them will quietly get dropped.

And the value of something that's watching all of it for you grows at the same rate.

An architecture that captures the meaning of operations works the same whether you're on one screen, ten screens, or a spatial canvas. Because it's not capturing pixels — it's capturing what you're doing. The richer the user's environment gets, the sharper the output gets.

Add a monitor. SHOGUN gets sharper. That's not a metaphor. It's the structure.

7. Why this is an OS for individuals

Worth being explicit about who this is for.

SHOGUN is not an enterprise product. At least not at the start.

Org-wide knowledge integration, enterprise search, cross-departmental data — that's a huge market with massive incumbents already in it. And the problems there are not technical. They're political — permissions, approvals, change management. That's a different fight. Not mine.

What I'm building for is the AI-native individual.

Founders, builders, researchers, creators, traders, operators of agents. Whether they're inside a company or not, people who create value with their own judgment and attention. People running multiple projects in parallel. People delegating work to multiple agents. People who already know, viscerally, that the bottleneck is the capacity of their own brain.

This group is growing. Companies are getting smaller. Individuals carry more leverage. One person plays multiple corporate functions at once. AI has accelerated that.

These are exactly the people who need the first brain rebuilt.

Inside an organization, the org still absorbs part of the cognitive load. Assistants, peers, reports, managers — someone catches what falls. Operating as an individual, no one catches what falls. You either do it yourself or hand it to AI.

That's the first market. That's where this lands first.

8. Why now

Last point. Why do this now.

The technology converged. Capturing context passively, at usable accuracy, at acceptable weight — that's now possible. Storing it as meaning, at consumer-product cost — possible. Running agents on top of it, reliably — possible. Standards for agents to talk to other agents and other systems — emerging.

All of that arrived in 2026, simultaneously. If any one piece had been missing, this would still be too early. A year ago you couldn't build this. A year from now, the first-mover window is closed.

And the demand side is ready too.

There are people who want to delegate to AI but can't quite manage to. The reason is simple: handing context to the AI costs about as much as just doing the task. "Here's the background, here's the tone, here's the audience, now write it" — every single time. That's not actually saving anything.

There are people trying to run multiple agents and feeling their head explode. Agent A did something. The result needs to go to Agent B. The human is the bridge. That's not "delegating to AI." That's "doing the work of delegating to AI."

There are people drowning in tools and giving up on integration. New AI tools every week. They try them. They like them. But integrating them into the existing flow is too much, so they quit. Tools accumulate. Work doesn't go down.

All of these people are waiting for the same thing. Not a second brain. A redesign of the first one.

9. The conclusion

Work is no longer happening in your head. It's happening on the screen.

Second Brain was the idea of supplementing the biological brain. It was beautiful, but the assumption is outdated. The biological brain is no longer the center of work. The center is on the screen. We don't need a supplement to the brain. We need to rebuild the brain itself, on the screen.

It's not search. Search is a tool for remembering.

It's not recall. Recall is a passive feature.

It's something that drafts, observes, proposes judgment, acts. And it has to be built at the OS layer — not as another tool. Because the layer running underneath every operation in your work is, in the truest sense, the real context.

I'm dropping the kind word that keeps the human as the protagonist.

The human can step out of the lead role. Become the commander making the calls. The execution is absorbed by a new layer below. That's how you promote everyone out of the work.

Build the brain on the screen.

That's SHOGUN.