AI Brain > AI Agent

The “AI brain” seems to be the solution that more and more companies are coalescing around, but I'm not sure we're at the end state yet.

Read what Tobi of Shopify, Jack Dorsey of Block, and Ramp have written this year and you'll see that they're converging on the same playbook: centralize your AI efforts into one growing organism. Shared tooling, public iteration, work in the open. Tobi's Learning on the Shop Floor post is the clearest articulation — Shopify's River agent only works in public Slack channels (no DMs allowed), the whole company watches each other prompt, and the merge rate went from 36% to 77% in two months — not from retraining but from humans writing down what River should have known. Every team's taste flows into the agent.

So what is an “AI brain”? Is it a chat bot in every channel like River? Or a Mac OS X app like Ramp's Glass, that comes out of the box ready to work with everything logged in? Or is it Notion's Hub, with agents pulling context from your workspace and connected apps to act on your behalf? I think we're too soon to describe the shape — it's emerging in the space we already live and use, as more of a set of capabilities orchestrating around agents, background runs, persistent memory, and a loop where every team's work makes the next team's agent better.

Jack Dorsey pushes the framing all the way: a company organized as an intelligence, not a hierarchy. AI at the center, people at the edges, the org chart replaced by a “world model” coordinating work in real time.

Here's where I get stuck.

Rarely in my career have I worked at a company that's 100% transparent on financials, HR, M&A. The best ones are open around product and dev needs, but past a point, you have legal, ego, and context overload. Maybe the information's not done yet, maybe the deals aren't done yet. Maybe you're setting people's hopes up about an acquisition that may not happen. Tobi's “everything in the open” works for Shopify because engineering-with-an-agent is naturally public-shareable. Not all work in a company is.

So I don't think the company AI brain is quite the right metaphor. What I keep coming back to is team AI brains. Each team gets a brain inside its own trust boundary, with the context and access it needs to do its job. The brain's job is then to provide an interface for other teams to communicate with it, schedule tasks, and hand off work. You end up with twelve to twenty team brains sharing access to systems of records like GitHub or Notion.

A bunch of startups are starting to look like this. Paperclip is building the control plane. My own (Valet) is here. Type is going there too.

Even assuming we end up with team brains, the open questions are big. How do roles and permissions scale to a shared entity? How does human-in-the-loop evolve when you hand more trust to a central figure? How do you, the individual, embed taste and quality into this brain? What happens when you and your coworker disagree? Or worse, one sabotages or attacks your brain through a supply chain attack? Does it become a Zombie brain!?!

And the shape itself is unsettled. Is the AI brain an all-in-one desktop app and service, a full operating system for your business (Glass, Cowork, Goose)? Or is it a piece of the puzzle that intersects with other surfaces (Notion, Glean, Claude)? We're still too soon to know, but as Agent Builders, we're the ones figuring it out.