TOKENINTELLIGENCEAll episodes ↗
TOKENEric and John discuss bottlenecks as a mental model, uncovering why constraints are leverage, not blockers. Hands-on Tool Time is with Zo Computer, a stateful, powerful, AI-enabled cloud computer.
In the second half of Episode 1, Eric and John tackle “bottlenecks” as a core mental model: why they limit system output, when to keep them on purpose, and how to fix the right ones without creating worse slowdowns. They share examples from product development, content quality control at scale, and how the youngest child changes family life.
In Tool Time, they go hands-on with Zo Computer, an AI-enabled cloud computer with state, plus agents and a real file system. Eric shares his screen to explore use cases like media management, hybrid search over local files, and remote development, ultimately questioning where the day-to-day value beats existing tools. Eric analyzes his entire history of blog post markdown files, and they conclude that running AI against physical files will be a big deal, but wonder if Zo is the right form factor.
Identify the real constraint and keep good bottlenecks: Focus on the true bottleneck, not the noisiest part. Optimizing fast stages is wasted effort. Some constraints (security, editorial review) protect quality and safety, so preserve them intentionally.
Fewer focused people beat swarm tactics: Small, targeted groups resolve bottlenecks faster than all-hands pile-ons.
Prototype fast, still ship with specs: High-fidelity prototypes unblock product velocity, but clear specifications prevent new downstream bottlenecks.
Save long-running AI work as real artifacts: Working against files and services with memory beats transient chats when your work is long-running or spans multiple sessions.
Files beat context windows: Hybrid search over a real file system is faster and more precise than stuffing giant context windows.
What uses cases the remote AI computer will really solve: Tools like Zo seem well suited when it beats local workflows on security (code/data never leaves a controlled environment), scalable compute (beefy GPUs/CPU on demand), or collaborative persistence (shared stateful workspaces, services, and logs that multiple people and agents can access).
Mental model: bottlenecks
Tools and AI
Studio (00:02.637) All right, now that we have solved AI FOMO. Yep. And started a company. We started a company. We read an 80 year old speech. We solved AI FOMO. Not sure what you came here for, but that's what you got. If you made it this far. Right, yeah. Julia keeps telling me. To both of our moms. Yes, well, and she's like, you can't do like a two hour episode.
All right, now that we have solved AI FOMO. Yep. And started a company. We started a company. We read an 80 year old speech. We solved AI FOMO. Not sure what you came here for, but that's what you got. If you made it this far. Right, yeah. Julia keeps telling me. To both of our moms. Yes, well, and she's like, you can't do like a two hour episode.
She's like, you can that's what she said? Yeah. OK. So is this multiple episodes? I don't know. We'll see. We have to see. Yeah, all right. We'll fix it in post. Great. OK. Two more things to cover in this episode. One is a brief discussion on mental models, and the other is tool time. We're going to talk about Zo.computer or ZoComputer. Yeah. Awesome. OK. Mental models. So here, we can show this to the video camera.
She's like, you can that's what she said? Yeah. OK. So is this multiple episodes? I don't know. We'll see. We have to see. Yeah, all right. We'll fix it in post. Great. OK. Two more things to cover in this episode. One is a brief discussion on mental models, and the other is tool time. We're going to talk about Zo.computer or ZoComputer. Yeah. Awesome. OK. Mental models. So here, we can show this to the video camera.
I guess maybe I could hold it up in front of me, like, my camera. The great mental models. So I actually mentioned Shane Parrish earlier. The knowledge project and brief summary. I'm not intimately, I don't have an intimate knowledge of, you know, all of the beginnings, but he got really into Charlie Munger and Warren Buffett of Berkshire Hathaway fame.
In-person participant 1 (01:00.549) I guess maybe I could hold it up in front of me, like, my camera. The great mental models. So I actually mentioned Shane Parrish earlier. The knowledge project and brief summary. I'm not intimately, I don't have an intimate knowledge of, you know, all of the beginnings, but he got really into Charlie Munger and Warren Buffett of Berkshire Hathaway fame.
And Charlie Munger has this, he calls it a lattice work of mental models, and he has this theory, or practical theory is probably a better way to put it, that if you get a handle on the core mental models that describe different fundamental parts of the world, that you can sort of tackle, you have the mental agility to tackle almost any subject or make a,
And Charlie Munger has this, he calls it a lattice work of mental models, and he has this theory, or practical theory is probably a better way to put it, that if you get a handle on the core mental models that describe different fundamental parts of the world, that you can sort of tackle, you have the mental agility to tackle almost any subject or make a,
How do you navigate the pace of AI disruption? This mental model helps you decode AI hype, catch cartographer bias, and avoid being blinded by the past.
Explore this episode ↗