# Develop at Idea Velocity - Jeffrey Lee-Chan, Snapchat AI Engineer · 2026-07-11 Jeffrey Lee-Chan (Snapchat) and Austin (CMUX) explain how to achieve 'idea velocity' by building parallel multi-agent harnesses that let one engineer direct 10–20 coding agents instead of becoming the bottleneck. Lee-Chan argues the key is separating 'Agent Orchestrator Managers' from specialized workers to prevent low-level implementation bias, using OpenClaw for frictionless Slack-based communication and CMUX terminals for real-time parallelization and unbiased manager oversight. He demonstrates two apps—an AI RPG with dice-roll mechanics and a multi-AI analysis tool—built with this stack, and discusses token-burn tradeoffs between models like Codex 5.3, GPT-5.4, and Minimax. The episode also covers staging environments to avoid doubling token usage, the benefit of persistent memory and contextual guardrails, and the importance of pushing human interaction to the start or end of the workflow for improved parallelization. ## Questions this episode answers ### How does Jeffrey Lee-Chan's multi-agent harness allow one engineer to manage many coding agents without becoming a bottleneck? Jeffrey's harness uses Agent Orchestrator Managers to dispatch tasks to specialized workers, keeping human direction at the edges. Workers operate in parallel via work trees and tools like CMUX, while the orchestrator maintains contextual guardrails and persistent memory. This structure removes the engineer from the critical path, enabling them to oversee 10–20 agents and achieve 'idea velocity' with automated iteration. [4:11](https://aie.addtry.com/3a0a9c8f-f15d-42b2-bd54-07bf0e4f7b68?t=251000) ### Why does Jeffrey Lee-Chan prefer using OpenClaw over directly interacting with Claude Code for development tasks? Jeffrey chooses OpenClaw for specialization. He explains that Claude Code’s context quickly fills with implementation details like Claude MDs, skills, and MCPs, consuming about 25% of the available context. In contrast, OpenClaw focuses on the task’s goals, history, and spec, assembling a reasonable plan without low-level implementation bias, which leads to more intentional orchestration. [5:18](https://aie.addtry.com/3a0a9c8f-f15d-42b2-bd54-07bf0e4f7b68?t=318000) ### How has the role of human testing shifted as AI coding agents have improved, according to Jeffrey's experience? Jeffrey notes that his testing process has evolved significantly. A year ago, agents struggled with simple browser interactions like pop-ups, so manual testing was essential. Now, agents can handle many browser tests reliably. He recommends a layered testing approach—unit, integration, and visual—and continually reassessing which stages truly require human judgment, as agents improve every quarter or even every month. [6:04](https://aie.addtry.com/3a0a9c8f-f15d-42b2-bd54-07bf0e4f7b68?t=364000) ## Key moments - **[0:00] Intro** - **[1:03] OpenClaw Overview** - [1:31] OpenClaw's persistent memory allows brief commands like 'fix this thing' by recalling prior task context and Slack history. - **[3:17] Core Principles** - [4:11] Jeffrey Lee-Chan separates agent stack into Orchestrator Managers for goal context and workers for implementation to avoid implementation bias. - **[5:07] OpenClaw vs Claude** - [5:10] Q: Why not use Claude directly? — OpenClaw preserves goal context, avoiding 25% of context lost to implementation details. - [6:46] Agent capability for visual browser tests improved quarterly, now handling password pop-ups that previously required manual intervention. - **[7:43] Website Demos** - [8:46] Multi-model consensus answers via his analysis website outperform single-model responses for research. - **[10:01] Agent Orchestration** - [10:17] Using CMUX as a manager rather than coder prevents worker agents from falsely claiming success—manager context correctly closes a PR. - **[11:24] CMUX Integration** - [11:24] CMUX adds Claude Code Teams integration to visualize sub-agent terminals and SSH support for remote development. - **[12:33] Sandbox Setup** - **[14:52] Model Selection** - [14:53] Q: Which coding model do you use? — Codex 5.3 over GPT 5.4 because GPT-5.4 burns more tokens; falls back to Minimax when rate limited. ## Speakers - **Austin** (guest) - **Jeffrey Lee-Chan** (guest) ## Topics Multi-Agent Orchestration, Coding Agents ## Mentioned Agent Orchestrator (product), CMUX (product), ChatGPT (product), Claude (product), Codex (product), Docker (product), GPT (product), Gemini (product), Minimax (product), OpenClaw (product), Slack (product), Tailscale (product) ## Transcript ### Intro **Jeffrey Lee-Chan** [0:01] Workshop setup, so, um, there are some people who kind of have, like, never installed OpenClaw, and I want to kind of get them set up. And then the other people, like, want to try an experimental thing, which I haven't even fully gotten working, but I think for more advanced people that'll be cool. Um, to get, like, a sort of staging environment and run to OpenClaw. So I think if I'm not, you know, too busy helping new people set up, then I'll try to help people, like, finish debugging the last part, or maybe someone else will get it. Then I think, uh, 10 minutes Q&A, 10 minutes some more advanced topics, and then just more Q&A. And, uh, probably after every 5 minutes or so, we can do a bit of, uh, a few minutes of discussion. So I'll just set some timers to make sure I don't go off track. Allright, let me start. Allright. Can everyone see this presentation? ### OpenClaw Overview **Guest** [1:08] Yes. **Jeffrey Lee-Chan** [1:08] Allright, cool. **Guest** [1:10] Yep. **Jeffrey Lee-Chan** [1:11] So this kind of describes, like, my setup. Um, I'll just keep it really high level for this part where, when everything works, you know, maybe like 70% of the time, you describe your task. And then what's nice is OpenClaw will have, like, context and memory about what you're talking about. So, for example, like, let's say I'm like, "Allright, fix this thing." It's a very brief Slack message, but OpenClaw can remember, like, what I've asked it to do before. So, for example, I'm working on this thing I'm calling a skeptic agent, and it's a custom code review thing. But if I'm like, "Allright, fix the skeptic agent," or whatever,right, it kind of knows what that is, which is nice. You don't have to re-explain a lot. Um, I use multiple agents with work trees. That's very key for parallelization. Uh, CI. Under integration. But you can kind of see you can talk to it and be pretty brief and then, you know, usually get reasonable outcomes, but sometimes you'll have to debug it. So then it's like, allright, there's some timeout thing, I'm like, give it a longer timeout,right. And then I asked it here, like, uh, you know, what are the priorities and what we're working on. And I'm reading it here. And then it might have just said, like, I just made code for it,right. So another interesting thing about this is you can kind of see from, like, um, the types of responses I give, they're not that, like, um, special. As in, I actually think an agent could replace me. So that's kind of what I'm working onright now to, like, get things even more autonomous. You know, and you'll see this with LLMs, like, you're working with them and they'll be like, "Hey, I've got this thing ready." And I'm like, "Okay, well, run the test,"right. Allright, so I'll show the next slide now. Oops. Okay. We'll go into some conceptsright now. Like, what is OpenClaw? So I think what's cool about OpenClaw is it's not just about a particular repo or the code, but more of the concept. So, you know, I've got this cool super diagram, but first concept is frictionless communication. ### Core Principles **Jeffrey Lee-Chan** [3:36] So I know we have other things like Claude cowork or whatever, but I think they still have a similar concept where it's like, can you, like, easily talk to your AI versus like, you've got a, um, remote desktop into your computer or you have to go sit down. So that kind of enables, uh, certain really higher level of velocity. Um, then I think the central axis is very important. So, you know, there's a spectrum. Like, I kind of go, like, a little more cowboy and I give it access. OpenClaw has these Agent Orchestrator Managers that manage your workers for you. Uh, or sometimes when I want more personal control, I use CMUX terminals. So CMUX is like a terminal program, but it's pretty good for parallelization and, um, AI development. Then I use the open source framework that I forked Agent Orchestrator to do the workers. And then once it gets to here, this part is, like, not really controlled by me as much anymore. Um, so I'm calling them managed agents, but basically, like, you have a worker that runs Claude code, and Claude code itself can run agents, and those can have sub-agents. So once you get to this Claude code part,right, this part is not exactly my stack, just stuff I use, but this part of the stack I'm changing a lot more. Um, I'll, I'll pause here to see if anyone has any questions or thoughts. **Guest 2** [5:07] Um, can you hear me? So I, I have a question. So. ### OpenClaw vs Claude **Jeffrey Lee-Chan** [5:09] Go ahead, yeah. **Guest 2** [5:10] The, the real question is, you know, why, why use OpenClaw versus directly use Claude? **Jeffrey Lee-Chan** [5:18] So yeah, that's, that's definitely a good question. Um, the reason I use OpenClaw is specialization. So when OpenClaw makes a decision, I want that context to be more about, like, the spec or the goals or, like, the history of what I want in the task, rather than the code. And you can just imagine this,right? As soon as you open up Claude, it reads Claude MDs, it reads skills, it reads, um, MCPs. A lot of those things are sort of independent of, like, the actual task. It's more about how to do the task. Um, so imagine, like, 25% of your context already taken up by implementation. Versus, like, when you think about OpenClaw, you're like, okay, I want to think about exactly what I want to do and how it relates to all the other Slacks Jeffrey has sent me in the last two weeks and put it all together and give me, like, a reasonable spec. Does every single thing a player does, but it does it without a browser,right. So you could use MCP, which usually is JSON or whatever, or HTTP. Doesn't really matter, actually. And then the final thing is the browser test. So some things like are visual. I mean, maybe you could even have CSS tests or JS tests,right, and that kind of goes here. But then the final thing is visual. So I would recommend you, like, try, um, similar approaches to this and be like, okay, like, which problems truly needed that human or not. Um, and I think what's kind of cool is, like, every quarter or maybe even every month, like, it improves. So before, like, I had to really manually test a lot because, like, you know, this worked fine,right, but these things didn't work that well with agents, like last year, like a year ago, even six months ago. But now, like, um, agents are pretty good at, like, nailing down a lot of browser tests for me. Versus, like, when I started, my agent would have a lot of problems, you know, finding a pop-up and entering a password. Now, no problem. Um, that was a little long, but does that kind of address the theme of your question? **Guest 2** [7:21] Yeah, I think this is good, good. Thank you. **Jeffrey Lee-Chan** [7:23] Okay. Just, just remind me for all these Slash commands. I'll put, um, links here instead. Allright, cool. Anyone have any other questions related to some of that stuff? Allright. If not, I will go here. ### Website Demos **Jeffrey Lee-Chan** [7:43] Allright, so here's, uh, two websites that I've built. While that loads, I'll show this one. So this is like an AI RPG. Um, what's cool about this is you can I've got a default campaign, for example. Um, I can pick some options. I've got an avatar. And, uh, what's cool about this is it builds a custom world for you that reacts to you. So you can, like, kind of say whatever you want, do whatever you want. You'll get a story back. Um, the main difference between playing this versus, like, um, you know, you could just always go into Gemini chat or, or ChatGPT or whatever is that, um, I have, like, a D&D system. So if you play your own, like, games or novels, like, you kind of just win too much. Versus, like, with this, you'll actually do, like, dice rolls and be like, you know, did the person actually, um, did you actually succeed in your action or not. Allright, so while this is going, I'll also, um, show this thing. Yeah, so this is funny. Um, this website is like a multi-AI analysis website. Um, so what I found was, like, whenever I was doing research or, or whatever, I would go to multiple models and I would be like, what's the answer? Then I copy and paste them all. I put them into one model. Um, so pretty simple concept, but this does it for you. So usually I like these answers better than, like, asking one model and I'm like, okay, how do I do this workshop? So I can be like, okay, what's the status? Um, the latest PR now. And, um, what I like about this is it has that sort of vertical tab type of thing. Because when you go horizontal, it's really easy to lose track of your tabs. Um, but we've got notifications too. So when this thing is finished or this is finished and I need to look, it'll give me a notification. And I can just focus on clearing the notifications. So here, um, the way I'm using these terminals, usually it's more like a manager rather than a coder. ### Agent Orchestration **Jeffrey Lee-Chan** [10:01] Um, and it's kind of interesting, but it gives you a certain benefit where I feel, I feel like the worker is not biased anymore. So when I, when I code with these directly, um, I usually feel like there's a bias where it wants to say things are really working or whatever,right. Versus, like, you know, um, and I'll show this. One sec. If this had been working in PR 294 by itself, I think it would have been like, this PR is amazing. Like, we got to merge it,right. But then this one was like, no, like, there's another PR that should supersede it and probably we should just close this PR,right. So that's kind of the benefit you get where the manager has a different context, um, than the workers. Uh, let's see. Allright, so this is compacting whatever. So it'll take some time to complete. Don't want to wait for it, but when it completes, you see these notifications and I think my efficiency has improved a lot with CMUX. Um, I think is Austin or someone from CMUX here? **Austin** [11:05] Yeah, I'm here. What's up? **Jeffrey Lee-Chan** [11:07] Cool. Yeah, so Austin is, uh, one of the guys behind CMUX. Um, you know, big fan of it. And I'm not, I'm not paid to say this or anything, but big fan of CMUX. Uh, I've been using it a lot. I don't know, you can spend a minute if you want to talk about anything cool about CMUX. ### CMUX Integration **Austin** [11:24] Yeah, so, yeah, I'm one of the creators of CMUX. Uh, really happy to have Jeff as one of our power users. Um, and also, like, kind of endorsing it for free. Um, but yeah, lots of, uh, yeah, lots of things that we're shipping. We shipped, uh, Claude Code Teams integration. So if anyone is using Claude Code Teams and wants to actually see what the sub-agents are doing, it'll automatically spawn terminals for that. Um, we also shipped, uh, CMUX SSH. So if you do any SSH work, um, you can use our native CMUX SSH to, you know, do your Tailscale, et cetera. Um, to other computers, you can even use it to run your own OpenClaw, um, in your Mac minis. Um, but yeah, just, uh, one of the creators. Please feel free to email me if you have anything that you think is a bug or any feedback, feature request. Um, I'm always available. And yeah, I just love seeing people using CMUX and always want to take as much feedback as I can to, like, get them as, get them. **Guest 3** [12:33] Okay, so just download that. ### Sandbox Setup **Jeffrey Lee-Chan** [12:34] Yeah. They actually have an app too, so the app might be nice. **Guest 3** [12:38] Yeah. Okay. And then use that to help me get back in and set, set up. **Jeffrey Lee-Chan** [12:44] Yeah, exactly. Yeah. **Guest 3** [12:45] Okay. Okay, cool. Thanks. **Jeffrey Lee-Chan** [12:47] Allright. Just using the sandbox. It should not. Where would you, um, have you, have you run into any troubles, Aaron, about sandbox token usage? **Guest** [12:57] Uh, no. I'm just, I'm not technical at all. **Jeffrey Lee-Chan** [13:01] Oh, yeah, yeah, yeah. **Guest** [13:02] I have OpenClaw installed, so I'm just. **Jeffrey Lee-Chan** [13:04] Okay. **Guest** [13:04] Very newbie questions. **Jeffrey Lee-Chan** [13:06] No, no, no. **Guest** [13:07] But I appreciate you being friendly. **Jeffrey Lee-Chan** [13:08] No bad questions. All good. Um, yeah. Uh, no, the sandbox wouldn't do that. Um, I think if, if everything's working well for you, you know, keep using the sandbox. For me, I just, you know, don't use the sandbox and so, so far nothing bad has happened. But I think if I had an external bot, then I would use the sandbox too. **Guest** [13:28] Changing the model, testing a new model, like, it'll go down and then I have to figure out how to get it back up again. So I think I want to set up the sandbox because. **Jeffrey Lee-Chan** [13:35] Oh, sorry, sorry. Um, okay, wait. There's two kinds of things. Okay. The sandbox that most people talk about means, like, they run it in Docker or they, like, have it running in an isolated part of their system. But the other one was the one where I made the staging instructions where I was like, oh. **Guest** [13:51] Yeah. **Jeffrey Lee-Chan** [13:52] Okay. So. **Guest** [13:53] That's what I'm referring to. **Jeffrey Lee-Chan** [13:54] That could double your token usage if you send the same work to both of them. If you have two of them,right. Um, but I think the way I would use it is not like that. Like, I think I would do local development and then I would run integration tests on the, um, sandbox or staging one,right. So I have two of them. And then once, like, everything's good, then I would merge the code and deploy to the production one. Um. **Guest** [14:16] I see. **Jeffrey Lee-Chan** [14:17] So it wouldn't double your usage, but it would definitely increase it. But then the benefit is you might have some more reliability. So. **Guest** [14:23] Yeah. **Jeffrey Lee-Chan** [14:23] I, I made some instructions and I was kind of in the middle of setting it up, but then, you know, it's like competing with the other one. So I got to debug it. **Guest** [14:30] Yeah. **Jeffrey Lee-Chan** [14:31] But I think, you know, it's, it's worth trying out though. So I add that for more advanced people to give it a, uh, give it a try. **Guest** [14:38] Cool. Thanks. What are. **Jeffrey Lee-Chan** [14:39] Or even if you're not as advanced, if you want it, you know, you can try it. **Guest** [14:42] Yeah, I'm definitely going to play around. But what, uh, what model do you use? Like, do you have a something that, like, a default orchestrator model and then use, like. **Jeffrey Lee-Chan** [14:52] Yeah, so. ### Model Selection **Guest** [14:52] Say for this model. **Jeffrey Lee-Chan** [14:53] I mostly use Codex 5.3. Um, I found, I found, uh, GPT 5.4 to just use more tokens. Um, and even 5.3, like, like, I, I just get destroyed all the time,right. So I might have to adjust. **Guest** [15:08] Yeah. **Jeffrey Lee-Chan** [15:08] Um, so basically I use this until, like, this is getting low and then I just switch to Minimax, which is not as good, but it kind of gets the job done. And then. **Guest** [15:18] Yeah. **Jeffrey Lee-Chan** [15:18] This is more about money than, like, preference, but some, some work I can still kind of just do, um, here. So I don't always have to. --- This library is powered by PodHood (https://podhood.com), the podcast website platform.