Lenny's Podcast: Product | Career | Growth
00:39:00 · Claire Vo
[00:39:00] S1: Well, and maybe this is why I feel like OpenClaw is just purpose-built for Claire Vo, which is been a manager and a leader for a long time, I know how to onboard an employee, I know how to give them a good role. I know how to set them up for success technically and inside an organization. Managers, this is your moment. You can design your OpenClaw and design your team of OpenClaw's using those organizational skills that you've developed over your career. And then I think about being a parent, I think I posted this, I said, "What a funny time where we're parenting and also literally writing markdown files for souls." And what I think about that, again, I don't personify my agents in my intellectual mind, but these large language models are trained on human text and are optimized to interact with humans that are social creatures. And so I do thinking about ceding identity in a way that is aligned and helpful and useful and then interacting with your agent in a way that is polite and proactive and collaborative, it's not because of the AI overlords, although if we have the uprising I was very nice, it's because I think you get better outcomes from it. Just as you would get better outcomes from a human system showing up with respect and organization and the transparency you need and the privacy you need, you're going to get better outcomes from these agents that way, I think.
00:40:33 · Lenny Rachitsky
[00:40:33] S0: I also say thank you often. And even though it's burning tokens and trusting money in theory, just in case. Okay. Talk about the different claws. So maybe even zooming out, it feels like a big unlock for you with OpenClaw was realizing you shouldn't just have one, you should have many that have very purpose built. Talk about that unlock and then just what the actual claws you have running are and what they do.
00:40:59 · Claire Vo
[00:40:59] S1: Part of I think where people stumble with OpenClaw is they read about OpenClaw is running my business, and they think they can throw any task at a single agent and get great results. And then they get really frustrated. I don't know if you've had this experience, my OpenClaw sometimes forgets what we talked about yesterday, even though they have a memory, it sometimes loses access to my email and asks me to reauthenticate or says it can't access a file that it can. And this really comes down to one concept that I think we're all really familiar with, which is context overload. We've seen this with ChatGPT, if you're using Claude Code for coding, the longer you go and fill out the context window, the harder it is for the agent to do a good job at the task at hand. And so you can manage your context window and OpenClaw does by compressing history at the end of a conversation or starting a fresh session, but I manage context windows even more efficiently by sectioning off which tasks go to which agent. So even between Polly and Finn, my work assistant and my family assistant, they're both doing scheduling and calendaring and email and admin stuff, but Polly has enough to worry about with the work stuff that I don't need her thinking about the kids' soccer schedule as well. And Finn has enough to think about that I don't need her replying to emails in my work inbox. And so that's where I started to really feel, oh, I would hire different people to do this job in real life, so I'm going to, quote unquote, hire different agents to do this job in my agent team. And then once I got that unluck, boy, did I start creating agents? So now we have Polly, Finn, Max, Howie, Sam, Kelly, Holly, and Sage and Q, nine. And then my husband has one over here. He takes a very different approach to naming. His is named Martron1000. So naming your OpenClaw is a very fun part of this process.
00:43:15 · Lenny Rachitsky, Claire Vo
[00:43:15] S0: Some people may think this as AI psychosis, having nine OpenClaw's. [00:43:23] S1: It's not psychosis. Let me change your mind here. Let's just make it a little bit simpler for you. [00:43:30] S0: I don't actually think that, just to be clear. [00:43:31] S1: I have nine Slack channels that I do my work in. I wouldn't put it all in general. I have nine Slack channels with my team and my marketing team's in one, and my sales team's in another, and my dev team's in another, and my development team does not care what was posted on X today, and my podcast team does not care what's in the ChatPRD sales pipeline. And so we don't have to make it weird. We can make it very practical, which is channels and different areas for different lanes of work and they intersect when they need to intersect. And otherwise we don't bother our colleagues with it. [00:44:09] S0: That is such a good way of framing it. And I never thought of it, the limitation is context and memory and having it focused and not forget basically.
00:44:09 · Lenny Rachitsky, Claire Vo
[00:44:09] S0: That is such a good way of framing it. And I never thought of it, the limitation is context and memory and having it focused and not forget basically. [00:44:18] S1: Well, and we're all coming out of many, many years of remote working and we know how context fried we all get in something like Slack where we're watching, "Oh, this customer conversation popped up in this channel and then I got a DM over here," and imagine if that was all happening in one stream of consciousness thread and you had to process it, that would be very hard. And so I just take that same mental model and I say, "Sam's not going to be able to process, by stream of consciousness, business and know when he needs to come in and when he doesn't. So I'm going to give his own quiet room and we're only going to talk about the stuff that he cares about there." [00:44:55] S0: Imagine your advice is start with one, play around and then add more. Don't just jump in in all the different use cases. [00:45:00] S1: And then add more. [00:45:00] S0: Cool. [00:45:01] S1: And then buy a bunch of MacBooks. [00:45:04] S0: How do you do multiple claws? Do you need a separate computer for each one?
00:45:04 · Lenny Rachitsky, Claire Vo
[00:45:04] S0: How do you do multiple claws? Do you need a separate computer for each one? [00:45:07] S1: You don't. And in fact, I have a Mac mini over here. It's very classy. It doesn't have the lobster headband. It has a piece of painter's tape on which I've scribbled Polly and Crew. And so if you haven't watched my How I AI episode with Jesse Genet, she's the one that got my mental model really unlocked around separate machines. And her concept and my concept now is if I'm okay with every agent on the machine occasionally going into each other's space, occasionally going into each other's tools, occasionally reading each other's docs, they can all live on the same machine. So you know what? I don't necessarily want Sam the salesperson looking in GitHub, it's not his job, but it doesn't cause me any heartache as an employer if my salesperson has access to reading GitHub. Not a big deal. And so they can all live on the same machine. But Finn, for example, my family agent, I'm going to move to his own machine. He has no business knowing what's going on in my work, so I don't need him in there. And I'm going to add my personal email address, which I haven't added so far in this workspace, and he's going to live physically partitioned from the rest of that team. Just like people walk around with their personal phone and their work phone and they say, "These are just different things that I do different work in," and keep those physically partitioned so that they literally cannot cross boundaries.