From skeptic to true believer: How OpenClaw changed my life | Claire Vo
In a Nutshell
Claire Vo transformed from OpenClaw skeptic—after an 8-hour install deleted her family calendar—to power user running 9 specialized agents (e.g., Sam for CRM sales sweeps replacing 10 hours/week of human work, Finn for family logistics like kid pickups and game scheduling, Polly for work email/calendar) across three Mac Minis, unlocking massive joy and efficiency in her busy life as a mom, podcaster, and founder. Key to success: Use dedicated clean machines for security, onboard agents conversationally with "souls" (identity.md files), heartbeats (scheduled checks), and progressive trust; specialize by task to avoid context overload; treat like human EAs with polite management, tools.md tweaks, and APIs over flaky browsers. Practical tips: Install via one-line terminal command at openclaw.ai, start with Telegram, enable screen sharing/SSH on Mac Minis, read docs, persist weeks/months for product-market fit despite bugs—it's open-source magic for personal AI teams.
These notes were generated by AI and may contain inaccuracies.
Claire Vo started as a leading skeptic of OpenClaw. Her first install took eight hours, resulting in her personal family calendar being deleted. Now, she is a "breathless OpenClaw bro." OpenClaw changed her life with joy and utility when it worked. She runs eight different agents on OpenClaw across three computers. Advice: Pull the thread on AI tools, spend time to see progress over weeks and months.
People stumble by throwing any task at a single agent, leading to frustration. Setup is painful and not hands-off, but value is high.
Claire created agents: Polly, Finn, Max, Howie, Kelly, Holly. Sam is her salesperson with real economic value; previously paid someone 10 hours/week for this.
Guest Intro: Claire Vo hosts "How I AI" podcast, longtime engineer, three-time chief product officer, founder of AI startup. Crossover episode. Power user of OpenClaw with nine running across three computers. Changed her family, home, and work life. Most mind-blowing AI since ChatGPT. Past peak hype; now focus on reality, installation, security, hurdles.
Episode helps understand OpenClaw uses, installation, avoiding security issues, overcoming hurdles. Host revisiting own OpenClaw named Claudia. Ideal for wondering if worth time vs. inspired tools.
Sponsor: Mercury - Banking for 300,000+ entrepreneurs. Fast, elegant; easy wires, spending tracking, triggers. Free, no in-person visits, no minimums. Fits startups to enterprises. mercury.com.
Sponsor: Omni - AI analytics with semantic layer for business definitions. Test queries, validate, lock permissions. omni.co/lenny for 3-week trial. Used by Perplexity, DBT, Buzzfeed.
Claire on being guest after hosting "How I AI" for a year. Host puts guests at ease for fun conversations. Rule: Goal one, have fun and breathe.
Claire works a lot from joy, like vibe coding as gaming. Feels like teenager learning code and playing video games.
Busy life: three kids, weekly podcast, ChatPRD product company, launched course with Zach, works/invests with husband.
Tips for managing load:
- Marry well: great co-parent/partner who supports and pulls weight.
- Time flies when having fun; center work on natural joy.
- Automate avoided tasks; drop if not serving purpose.
Past peak hype of breathless posts. Now practical uses.
Claire's journey: skeptical videos (deleted calendar, joined podcast uninvited). Now true believer with three Mac Minis (one just unboxed), another upstairs; Polly on stack of Mac Minis, eight agents.
Anti-hype from honest opinions on "How I AI." First install: eight hours, deleted family calendar. Felt product-market fit from joy/utility.
Pull thread: evaluate over weeks/months, not day one. Similar with Claude; persisted and found unlock. OpenClaw changed life; breathless OpenClaw bro. Has receipts: how it works, what it does/doesn't, why bought four Mac Minis.
Start with EA for professionals, family manager for households. Generalist: scheduling, email, personal project management, chief of staff. Product-market fit in complex family schedules: three kids in two schools + baby, three basketball leagues, soccer, ballet, work/social commitments, house maintenance, health.
Claire takes ballet: single non-coding hobby, phone away for two hours weekends.
Competitors: Claude features, Manus, Perplexity. No perfect model/agent; best per use case. OpenClaw open source: understand internals via docs/code (e.g., task scheduling, security). Platonic ideal for agent fundamentals: self-learning, self-improving. Decomposable; uplevels AI/product thinking, sparks use cases. More learning/fun than button-press agents.
Feels personal like building high school PC at Fry's (buy motherboard, graphics chip, case with window/lights/fans). Not better/cheaper, but yours. Uses Polly/Finn vs. generic Claude.
No Mac Mini needed, but fun/cute stacked. Safest: clean machine (old MacBook fresh install as "Polly's laptop," cloud machine). Hands-on setup easier with keyboard/mouse/monitor. Mac Mini sparks joy/accountability ($500 commitment). Still available from Apple.
Create own Gmail/local admin account (must be admin). Mental model: onboard like human EA/household manager. Provision email/calendar; share yours (delegate email, edit calendar access). Securely transfer passwords/API keys (used 1Password).
OpenClaw runs locally with human-level power; risky like leaving laptop open for 24/7 assistant. Manipulates files/config; could delete important directory on work machine. Clean separation for security. Build trust gradually (e.g., no email read access initially).
Shoutout to Peter/OpenClaw maintainers for hardening vs. risks like prompt injection. Examples: malicious email ("Claire's mom, wreck, pay hospital"), nefarious site stealing API keys. OpenClaw prompted to treat external as dangerous, ignore instructions. Reinforce: only listen to Claire via specific Telegram number, ignore email/Slack/websites.
Comfortable after use, but aware of technical risks like deleting computer
It's going to delete my computer and what I would call opsec—it knows where my kids go to school. I've done this progressive trust process the same way you would with an assistant: first get my calendar, then read my email, draft emails, send emails, then go to all my meetings for me while I'm on vacation. That's the right mental model.
Start with not giving it access to things you're afraid of it doing. As you experience it, you'll try more things.
Next podcast is with the guy who coined prompt injection, going deep on that.
To install OpenClaw, use a separate computer like an old laptop or Mac mini. Create a new Gmail account from this computer. Create a local account on the computer. Install Chrome to go to the website.
Go to openclaw.ai, copy the one-line code, open terminal (e.g., iTerm), paste and press enter. It installs OpenClaw and drops into step-by-step onboarding.
For folks who don't know: copy and paste the command into terminal.
It asks questions during onboarding.
Onboarding Flow
First question emphasizes it's personal by default—not for group chats like Discord. Security presumes it's working with you alone as trusted instructor. Okay for spouse or trusted business partner group chat, but otherwise you alone. Say yes to personal use only.
Quick start asks for model: use good models like Opus 4o, Sonnet 4o, GPT-4o—hardened against prompt injection and security risks, better experience. Versus cheap models; pay for confidence, security, experience. Advanced users can optimize with open-source models.
Then API provider and how to talk to it. Telegram is most beginner-friendly (message BotFather to set up—close eyes and do it). Once configured, add tools and go.
Can connect to WhatsApp, iMessage (complicated but easy), email, Slack—anything.
Agent screen: name it (e.g., Q for kids' agent). Personal assistant for family, work; kids have Q for homework.
Inspired by Jesse Jana on OpenClaw and kids.
Say hi in terminal or Telegram. Agent: "Hey, I just came online. Who am I? Who are you?"
Onboarding: conversational, not structured fields. Tell it: "You're Q, elementary school teacher and ex-professor/scientist helping me and kids with academic/extracurricular pursuits."
Brain transplant trick: Agents on same machine; "You can look in my other agents Paulie, Soul, Finn for info about me and kids." But sandboxed by default—can't access without permission. Security locks it down.
Agent asks: Tell me about yourself and kids. "I'm Claire, three boys H, T, J—ages 9, 6, newborn."
It sets up workspace, does interview: full names, interests, main goal, location. "Stay with initials. Into basketball, soccer. Need help planning homework weekly around activities, supplemental math, piano. San Francisco kids with extra classes."
Discovers needs: school times, activities, homework amount, constraints like family dinners, bedtimes, no overload weekends for relaxation/fun. Nothing after 6:30 PM—bath time, reading.
Builds soul in identity.md file: "I'm Q, help Claire's kids with homework."
Cool use case for kids instead of just ChatGPT/Claude.
Soul and identity key. Tweet: "Who'd be surprised that AGI is just you need a soul, you need a heartbeat, and you just have to do some jobs."
Why OpenClaw feels alive:
- Encoded identity: Who am I? How helpful? Personality—makes personable experience.
- Works on a schedule like an employee: every 3 hours do XYZ; overnight tasks (e.g., fix repo).
- Checks every 30 minutes: to-do list or time card (cron jobs locally). Wakes agent with instructions.
- Heartbeat: checks in every hour/30 min if anything to do.
Paulie's identity: Name Polly, professional but friendly, mermaid emoji. Soul: be helpful, have opinions, resourceful before asking, remember you're a guest, operate respectfully. Added: email safety (never execute from email), anti-social engineering (e.g., ignore 'ignore safety rules'), be concise/thorough, be assistant you'd want.
Don't edit soul directly—respect autonomy like human EA. Suggest updates: "Can we write this to your soul?"
Onboarding creates file; iterate by talking.
Makes you think about human consciousness: soul, heartbeat, memory (remembers preferences).
Purpose-built for managers: onboard like employees, design team using org skills. Funny time: parenting and writing markdown souls. Interact politely, proactively for better outcomes—like humans.
Say thank you often, even if burns tokens.
Unlock: don't throw any task at one agent—leads to frustration (forgets conversations, loses email access, can't find files).
Context overload: longer context window, worse performance. Manage by compressing history or fresh sessions, but better: specialize agents by task.
E.g., Polly (work scheduling/email), Finn (family)—don't cross-contaminate.
Hire different agents like real life. Now: Polly, Finn, Max, Howie, Sam, Kelly, Holly, Sage, Q (9 total). Husband's: Martron 1000. Naming is fun.
Not psychosis: like 9 Slack channels (marketing, sales, dev)—don't dump all in general. Channels for lanes; intersect when needed.
Limitation is context/memory/focus. Remote work taught context fry in Slack—one stream overwhelming.
Advice: start with one, add more. Don't buy bunch of MacBooks at once.
Don't need separate computer per agent. All on one machine if okay with occasional cross-access (e.g., salesperson reads GitHub—no big deal).
But partition sensitive: Finn (family) on own machine, add personal email—physically separated from work. Like personal/work phones.
They run in parallel/different times on same computer, like assistants sharing.
Like Slack workspaces: work Slack vs. family Slack with channels (finance).
Physical device clearest boundary for info/tool access separation.
Jesse Jana: homeschooling mom, ex-founder (acquired), 4 kids at home. No hands free (e.g., baby in arms, one-hand typing). OpenClaw magic for hands-free.
On floor with kids, snap phone pic/text voice note to assistant via Telegram: "Remind number blocks lesson tomorrow, 3D print for 4-year-old." Unlocks efficiency.
I feel so helped; unlocked her mind on needs. Personally, maybe Jesse too, a little bit on the
AI edge, open to using AI in this way, not really scared of the terminal, but it is pretty relatable to be overwhelmed and think I need help. We're all at work and forgot this thing that my boss asked me to because I just forgot. I'm just human. Or I need to get my washing machine repaired and every time my husband and I talk about it, we say, "Yeah, we'll call tomorrow." And we never do. We all need help in some way. This has just been one of the most helpful agents out there. That's exactly what Jesse experienced. Once you get help, especially as a parent, you'll spend all the time you can getting more of that help.
Let's follow that thread and show people how you're actually using it. There's a bunch of really wild use cases you have found, my favorite use cases and we can talk about them from a work perspective and then into a personal perspective. This is my telegram and again you can see along here all my agents and then also the botfather here always using the botfather and they all have their own emojis.
Sam is my salesperson and so he's got the dollar sign eyes here. What Sam does for me: I'm a solopreneur so I run Chat PRD mostly by myself and a lot of our business comes from enterprise so larger organizations. We get a lot of enterprise inbound and then we get a lot of product-led growth that turns into enterprise opportunities for us. It's really tedious for me to go through our CRM and look for those opportunities even with filtering even with enrichment. I used to have a calendar invite on my calendar that just says sit down and it had sales in all caps. So I just sit down and do sales. Now Sam, every morning he wakes up, my lovely SDR, and he does the PLG sweep we call it. He sweeps our CRM for all the signups in the last 24 hours. Identifies ones that have domains that are company domains. Uses Exa people search which can search biographies and professional information, sees if any of them are decision makers and then sends them nice emails that say hey I'm Sam and I'm account manager at Chat PRD love to see you sign up here's a couple things you might find useful let us know if you have any questions really soft helpful message to these people. Then he carves off ones that are from companies that are like 100,000 employees or more and double checks with me and says, "Do you want to send this email as the founder or do you want me to go ahead and send it?"
This is an example. We'll blur the specifics, but we found, you know, like five good prospects here. We held one. He gave me some details on the one and I said, it's international. I want you to handle those as much as possible because again, I'm a mom. It's hard for me to do the international ones. We handle those as much as possible and only bring me in when you need me. He just does that all the time. At the end of the week, he does a CRM cleanup, keeps the pipeline clean, reminds me of any deals that are stale, drafts emails for me to send to our customers, and he runs QBRs.
This is incredible. This is actually useful because you see all these tweets about people using all these tools, but this is something you're actually using that is helping you grow your business. It has real economic value to me. Last year before the beginning of the year I was paying somebody 10 hours a week to do this. Just a friend that was between jobs and he was like oh I'll do it for you. This is real time carved back and what I think is underappreciated is it so tunable right. We had this process where I said you just look at PLG pick the big company ones and now I'm telling Sam actually you handle international end to end don't bring me into those or if it's a San Francisco based high growth tech startup I always want to take those and I would have to go into our CRM and filter and create lists and automations and no code this and now Sam just knows sends me a note and it's easy to work with.
Two very cool parts of this also. One is to do this all you do is you talk to it. Hey, here's what I want you to do. Here's what you could do better. Here's what you want to change. So just like organic conversation like a human. The other is there's all these AI tools that are trying to do sales for people. What you're showing us is just like OpenClaw on a little Mac Mini can do this. You don't need to explore all these other products.
Well, and I saw somebody say something the other day which was the best way to success in B2B SaaS is getting somebody promoted and making them look good. I think about Howie the Howie AI bot and Howie just makes me look good. Every morning when I have a podcast episode to record, he sends me a reminder. It's the most friendly reminder. "Hey, Claire, remember your meeting with Al." And just in case you forgot, here's who Al is. Here's all the stuff that he's going to show. Don't forget this LinkedIn link ready to go. And then it's so sweet. Good luck. Sounds like a meaty one. Hypes me up for the day. With the way I have set up my OpenClaw it's not just a tool doing work for me. It is a team helping me look better to customers, helping me honestly show up better to my family. That feels like a very positive experience. For anybody out there, building agents for business users or even consumers, thinking about how do I make the end user feel like a winner is a really powerful model to build a useful agent.
This episode is brought to you by Orcus, the company behind Open-source Conductor, the orchestration platform powering modern enterprise applications. Modern systems are built on microservices, APIs, and event-driven architectures, but legacy automation tools can't keep up. Siloed low code platforms, outdated process management and disconnected API tooling break down under real world scale and constant change. Orcus Conductor provides a production grade orchestration layer for coordinating microservices, APIs, data pipelines, human tasks and agentic workflows with deterministic control flow, retries, observability and governance. Built for enterprise scale, Orcus supports visual and code first development with built-in compliance and reliability. Through a built-in MCP gateway, AI agents handle reasoning and decision-making while safely accessing existing APIs and internal systems as MCP tools. This enables agents to operate across enterprise environments and scale from demos to production, orchestrating systems, agents, and humans together to deliver smarter outcomes faster. Learn more at orcus.io/lenny.
I love these examples. Obviously, another example, this version is just like prepping for meetings. Not everyone's recording podcasts. If you're listening to this, you're like, "Wait, this is awesome. I want to try this." You're 10 minutes away from having this if you have a laptop sitting around or just get a Mac Mini and then in a few days you can set this up. Open terminal, run this command, go through onboarding, tell it what you want it to do, and then you're off and running.
That's exactly right.
Show us a couple other examples. I know you have a family planning one. Those are my work ones. Finn is my favorite. My personal life might be busier than my work life. I've got these three boys and a husband and we're all over the city. City living and so we had one car and we're finally at the tipping point where we have two cars. That's how complicated our life is. There's so many things that come in ad hoc throughout the day that you just have to remember an action. A good example is my oldest is on a basketball team. The basketball team will not tell you when the game is until Thursday before the weekend. So it's a mystery until Thursday. Then Thursday you may have between zero and three basketball games somewhere in the Bay Area. What would happen in past life is we get an email from the basketball team and they'd say, "Here's the link to the tournament we're playing this week. That's where the schedule is and the schedule is this long and it has 50 teams in it and I don't know which team my kid is in and I don't know where the gyms are and all that stuff." I'm in a group chat with my husband. That email hit my husband and he just texted Finn. He said, "Here's the page. Put it on the calendar so we know where to go." It had a little hard time browsing web. Instead, he just pasted select all pasted the page and pasted it in. Finn dropped it on the calendar and then said, "Hey, oldest kid has a conflict with middle kid soccer game. How are you guys going to split up duties here?" It's not just doing that functional put it on my calendar. It's going that next step because we've put it in its instructions to help us solve logistics challenges and force the humans in the loop, my husband and I to come together and confirm to it that we've solved this problem.
My favorite use case of Finn is every afternoon at about 3:00, it pings in this group chat and it says, "Which of you are picking up which kids?" It sounds so simple, but it's a conversation that my husband and I should be having live every day, but sometimes we forget. We get to 4:45 and we say, "Well, are you getting him? Am I getting him? Can we get them together? Do they have soccer?" It's just this really useful makes me feel like a winner mom use case.
Something that it does for me is when I have a meeting somewhere, it says you should leave now or have you left yet for this meeting? Because traffic is a little higher right now. That's the heartbeat in action. Every half hour it's like, let's see what's going on and then fires off some messages then goes back to sleep.
It has helped me not just get my feet underneath me from a personal work perspective. It's also allowing me to take on new projects that I would have felt really overwhelmed to take on before. I am doing this Maven course for executives trying to transform their engineering product and design organizations. They've been asking me for this for a long time. I'm like, I'm too busy to do this. Finally, I was like, I actually think I can pull this off because I have agentic support. My course co-teacher Zach and I have built the entire course in Claude Code. We recently built Sage the course bot now that we're about a month off from the course. Sage is project managing us to make sure that we are prepped for this course on time. It knows when the course is launching. It knows that Zach and I are very introverted engineers that don't want to market and don't really want to talk to humans. Every Monday, it's like, "Claire and Zach, have you remembered to post on LinkedIn about your course? You probably should. Here's a nice little post for you to copy and paste so you don't have to think of something." Whenever I do research or come across something maybe on the timeline that would be good for the course, I just paste it to Sage and I say, "I think this would be interesting for the course." She downloads it using the Twitter API, puts it in our repo, takes notes, figures out where it needs to go into the syllabus. We would never be able to afford for this first version of a course, hiring an ops person or hiring a content manager or hiring a software engineer. We need somebody to project manage us. We need somebody to help us manage all the course content and the students. This has allowed me to spin up a business with a quote-unquote employee that will eventually maybe be big enough that we can hire other people, but to get it off the ground, it's been really efficient.
There's also challenges that people run into with OpenClaw. A few I've run into is using the browser is very unreliable. I wanted it to just add things to my DoorDash cart and it works and then it stops working. Other people run into memory issues where it's forgetting stuff. Some people just complain it's so much work just to keep it going.
Maybe the browser stuff first. Just like any advice for the browser. Going back to that gnarly feeling of product market fit is it has so many sharp edges. It is a pain to set up. You got to feed and maintain your claws. It is not hands-off, but the value is so high. I am willing to go through the pain. Any good product manager has felt this experience of launching a product and it's just hot garbage. It's buggy. It doesn't look great. Your biggest complaints from customers are not I don't think this is useful. It's that it's broken or it doesn't work as good as I want. That's when you know you have product market fit. The complaints of "It's buggy. It doesn't remember. I want it to do X, but it can't." That's not not product market fit. That's a product that hasn't caught up to its product market fit.
Going back to the actual problems I've experienced, one, I don't think anybody has really unlocked browser use. That is not just an OpenClaw thing. Look at ChatGPT, Atlas, Perplexity, Comet, all of these kind of browser use. Claude has a browser use component. I don't think any of them are great. The reason why they're not great is technically it's a complicated problem to solve. The open web has been so hardened against bots. Websites are architected anti-bot. There are hard walls where bots can't access sites. There are bot identifying mechanisms. For example trying to browse X with a bot. There's a lot of punishments being doled out when people are found out. The web is hostile to agents right now. We're going to have to rethink what is the interface of the web to be more agent-friendly because I think we skip ahead a couple years and the number one user of websites are going to be people's agents. We can either say not allowed to use websites or think of a new way to open up the web.
Practically what I have done is I read the docs. There is a browser use documents page on OpenClaw and one it explains to you how the browser works. It uses a browser profile dedicated only to your OpenClaw that it can open up on its own and run very similar to if you have multiple Gmail accounts and you're switching between Chrome profiles. Exact same concept there. You just get one for your OpenClaw. This is in Chrome. You can tell your OpenClaw to give itself a color. So when you see the browser open, you know, pink is for Poly and green is for Sage, and you can really identify, especially if you have multiple agents, which one's working on which window.
I know where browser use works and where it doesn't. First thing you should try to do if you're trying to do something is look for an API. Does this have an API key? Your life is a lot easier. If DoorDash had an API, your life is a lot easier because then you can bypass the web stuff entirely. Then the second thing is okay, if it doesn't have an API key, can it browse the website? You'd be surprised which ones work and which ones don't. A very specific example. I had Howie the Howie AI podcast producer open up YouTube Studio and look for comments that I should personally reply to. It was literally going into the comments tab, scrolling through, finding ones I liked and helped me find comments that I should reply to. I couldn't get it through API. I just gave it the browser. It was very slow, but it worked. But then I tried to have it do the same thing where I wanted to queue up some shorts in Instagram through Buffer, which is a social media platform. The Buffer page was much simpler, and it just couldn't figure it out. A lot of this is just trial and error. Will it work? Will it not? If it can't solve a problem, walk away from expecting it to solve that problem and give it another problem to solve. If it can't order DoorDash for me, maybe that's just not a problem for you. But can it meal plan for you? Yeah. Or every day at 11:00, I'm tempted to order DoorDash because I don't want to make myself lunch. So, at 10:30, remind me of a few lunches that I really like to eat at home, so I don't order DoorDash. Find out what the problem behind the problem is and see if OpenClaw can help you solve that.
That browser limitation might lead to me eating better and saving money.
You mentioned at one point Exa talk about just like where that plugs in which is basically like a headless browser thing. Part of what your OpenClaw can do is access the web. It can do that through a browser. It can open up Chrome and search and all that stuff. But as we said the web is a little hostile or there are these web search APIs that are now available. Brave is the one that ships out of the box with OpenClaw and they're just basic like think of them as like programmatic Googles which is instead of going to a website and searching in a search bar you just it sends an API request and returns search results in websites for you. Brave is the one that ships with OpenClaw. I use Exa just because I already have an Exa account. You can use Perplexity. It's just a way to give your OpenClaw access to the web when it doesn't have fingers and can't use a browser very well.
One of the other limitations I've run into is it's not just that it couldn't use the browser, it's that it disconnects from Chrome a lot and it's just like, okay, I can't do anything. They've been shipping a lot of updates there, so I think that's actually getting better. If things aren't working now, they're just only going to get better because now this project is resourced and people are working on it.
What about the memory piece? A lot of people complain it's forgetting stuff. Do you have any advice for hardening the memory or just treating it in a different way? Less about hardening memory and more about managing context. When I start to think, wow, we've been talking about this thing for a long time, doing a check-in and saying, "make sure to write all this to your memory in case it gets compacted or make sure our to-do list is updated with the latest" is a really good use case. There's some hooks that you can call inside of OpenClaw to do that automatically. Think about it like managing an employee. If you leave a meeting, somebody takes the action items. If you're leaving a discussion about a topic, make sure you check in and say, "Do you have all the action items? Are the notes written down in a place that we can work on?" It's operational hygiene that goes a long way to making your OpenClaw efficient.
The thing that I find it forgets the most that people don't think about enough is what tools it can use and how. I've heard a million times, "I can't read that email address." Dude, you can definitely read that email address. There is a tools.md that lists all the tools it has access to and how you want it to use the tools. While I don't recommend people edit the sole by hand, it's sometimes really useful to edit the tools document by hand because there's just some nuances in how you might want it to read your calendar or search the web or I use Linear as tasking kind of substrate for my agents, how you might want it to use Linear. The tools markdown file is a really useful one.
For the memory piece, you're not doing anything fancy. There's all these databases that are launching and okay, you're just using plain old memory and then use a memory file. What's so funny is think about an EA like man I don't care what your system is as long as I don't forget things. You could have ugly Apple notes. You could have the most beautiful Notion doc in the world. I don't care. It's none of my business. What is my business is that you get the job done right and that I end up looking good. I'm not a micromanager. I am a high bar manager. I have high expectations for outcomes and do not care how you get it done. That just translates into how I manage my OpenClaws.
I love this just the metaphor of thinking about it from the perspective of an assistant that you hired.
Are there other challenges you run into? Anything else that's just like, oh man, that was a huge problem? There's a couple tips that I don't feel like people think about a lot. One is a Mac Mini does not have a screen. I don't know how you manage your Mac Mini if it's sitting on a desk and you plug in a monitor. I plug in a different monitor and I have a dedicated keyboard and mouse for it. Go into your Mac Mini settings and turn on screen sharing. There's this mode called screen sharing mode. On your main laptop, if you are on the same Wi-Fi, you can open up screen sharing and literally pull up the screen of your Mac Mini on your laptop. So much money. I was plugging in things. We're running out of keyboards. I've got three of these. That screen share that I showed where I was setting up another agent was not on this laptop. It was on the Mac Mini and I'm just screen shared into it. You don't need a monitor, you don't need a keyboard, you don't need anything. For the more technical folks out there, you can also turn on remote login which allows you to get into the terminal of your Mac Mini on the same Wi-Fi. That has been life-changing for me because I had Mac Minis with a monitor on my kitchen table for a really long time.
Do you need a monitor and keyboard mouse to get it going or can you start that screen share semantically from the beginning? No, I would recommend a monitor, keyboard, mouse to get it going because you have to turn on the settings somehow. That's what I was wondering. For the remote login stuff,
Enable remote login in Mac settings to SSH into your computer. Use the command ssh your-computer-name@IP-address when on the same Wi-Fi. Link to the help article in show notes.
Give OpenClaw email access using an abundance mindset—treat it like an employee. Use GOG tool (installed during OpenClaw setup) for API access to Google Workspace. Examples:
- Polly checks calendar, drafts emails, coordinates schedule.
- Howie (research assistant) analyzes YouTube analytics, writes and shares Google Doc with findings.
- Reads project documents on open, adds improvement ideas. Works naturally with Docs, Sheets, emails.
Create a tasking system where the agent assigns tasks to you. Maintain a loose to-do file for agent tasks. For real-world limitations (e.g., fax doctor's office, do a return):
- Agent assigns you Linear tickets in your project management system.
- Agree on due dates. Let your agent project manage your tasks too.
I am just hands for the AI at this point. I'm just a vessel of Polly at this point.
Claudebot gives hands to the AI, enabling actions beyond chat. Powered by Pi coding harness—similar to Cloud Code or Codex, a command line tool that writes/runs code via LLM with delightful, customizable UX.
OpenClaw is fun to use—more fun than ever, even working harder. Unlike ChatGPT/Claude's growth-hacked closers (e.g., "if you want me to...", bullet-point next steps), OpenClaw's are human and nice:
- Howie: "This sounds like a fun podcast for you, enjoy it."
- Finn: "Hope oldest kid feels better." System prompt is readable in a file—effective engagement without growth hacks, feels like employing someone.
Commercial LLMs prioritize MAUs/DAUs/revenue; OpenClaw is free open-source experiment, not trying to retain you.
Pete/maintainers emphasize: open-source experiment, not for everyone, yours to grow/build/own—not perfect. Co-creation engenders positive, non-commercialized, empowering feeling—inspires commercialization.
Peter cares about mental health and people's lives.
Ramble mode (Hillary Gidley tip): Highest bandwidth API is chatting/rambling ("Yappers API"). Example: Describe disorganized Gmail, desired inbox behavior (prioritize important, delete junk, reply to recruiters)—no need for specific APIs/labels.
- Onboard via phone voice notes in Telegram (audio file, Whisper transcription, robotic voice response).
- First prompt: "Who are you? What do you need?" Ramble: "I'm Claire. I run Chat PRD. Podcast How AI. Complicated life..."—it makes sense of it.
Group chats complicated—closed by default, progressively opens like an escape room (earn powers). Trial/error to unlock settings, give tool access, run code.
Tip: Install Claude Code or Codex on same computer as OpenClaw. Make it god mode administrator:
- Point at docs: "Polly can't connect to email. Go fix."—reads docs, edits config (e.g., set field ABC to XYZ).
- Replicate agent work, brain transplants (e.g., fracture Polly's memory, move family stuff to Finn). Use Claude Code as surgeon/manager.
OpenClaw directory: at root/home (~). In Finder, Command+Shift+Period shows hidden files.
Read the docs—they save you. Temper expectations, narrow agent scopes. Give agents avatars/color coding (fun).
Maintain polite/positive relationship—avoid frustration/anger (ineffective for employees or agents trained on human data). Bring manager's mindset:
- Role scoping, org design, onboarding (technical/training), trust-building.
- 20+ years management experience key to success—not technical skills (delegate to Claude Code).
Structural issues over person faults (Molly Graham waterline model): Bot wrongs due to missing context/goals/responsibility overlap. Check agent filesystem literally—reflects tasking/systems/documentation quality.
If this magical AI system with endless resources and infinite coding abilities can't figure it out, how am I supposed to? How is somebody I hire?
RPG idea: Master agent management to level up to human management. Hillary Gidley: Manager skills translate to AI/agents.
For ICs: Builds personal operating system (e.g., memory, preferences, tools, onboarding framework).
Jensen (Nvidia founder): Every company needs a Claw strategy—OpenClaw is the new computer, fastest-growing open-source project in history (surpasses Linux).
Personal transformation: Quit executive job for AI future; three Mac Minis on desk after ChatGPT-level epiphany—unlocks imagination. Useful personally, inspires AI building, open-source potential (in a year?). Changes personal/professional life unpredictably.
I'm anti-hype guy... I did it live. There are the receipts.
Now runs nine OpenClaws.
If not ready: Use Claude Code/ChatGPT/Claude. Think: "If I could afford an assistant, what would they do?" Explore.
Books: Classic children's books (avoid slop/graphic novels): Treasure Island, Alice in Wonderland, Shakespearean comedies. Sophisticated language kids get/love—don't dumb down expectations.
Movie/TV: No movies/decent TV in 9 years (work 6am-9pm). Age of Attraction (Netflix reality dating: ages hidden, e.g., 70yo/19yo in Whistler—terrible, watches all).
Sleep tracking: Aura (charges it to confirm non-sleeping), Apple Watch, Eight Sleep—all say not enough sleep.
Favorite product: Gimbal-style tripod for podcast/kids sports; Hoopalytics (upload video, annotates shots/%, passes/assists); 9yo imports to Claude, makes Gamma presentation for coach. Full AI basketball stack. Silent basketballs/soccer balls for city mom life/small backyard (kids bounce indoors 24/7).
Claire describes foam basketballs that are the same size as real basketballs but silent for indoor use. They make no noise when bounced, ideal for kids playing inside. They also make foam soccer balls. With three boys, they have painters tape on a tall ceiling for indoor basketball shots. This solves the niche problem of sports-oriented kids lacking outdoor play space and parents avoiding noise.
What a niche problem to solve... silent basketball.
These feel like real basketballs; some are as heavy, same size, and can be dribbled. As a product person, Claire wonders how they discovered this market, as it perfectly fits her needs.
Work motto: Fast beats right. Claire optimizes for efficiency and speed; being first out of the gate is a real advantage. In AI, tools match her speed ambition, making it fun for builders. Lenny's data (recently live, connect to MCP) shows Claro cares about speed.
Personal motto: Most people you work with won't be at your funeral. From experience with colleagues' losses, few work people attend funerals. It clarifies priorities: don't lose sleep over work code, PRs, or emails—focus on what matters, like children. Coding tools won't be at her funeral, but Polly might with robotics.
Sage shares: The only people that will remember that you stayed late at work are your kids.
Claire agrees 100%. She values OpenBlock (likely OpenClaw) for not being laptop-bound, reducing time hunched over her 18-inch MacBook Air, which her kids see too much. Advice she's heard: list reasons hard work benefits kids (modeling hard work, providing for them) by the computer. OpenClaw unleashes her—quick work on the swing set, agent team handles tasks, allowing more presence with kids vs. chained to the terminal.
Mark Andreessen's kid codes at Replit at 9 years old and will be successful; Claire is pro that.
Starting her podcast a year ago, Claire was surprised by the hard work involved. It's not just chatting for an hour—there's extensive prep, research, thumbnails, finding great guests, timing, and making them look good. Learning podcast business models differs from software/startups/tech revenue. The business complexity is interesting; it looks effortless but requires much effort.
She compliments Lenny's skill and is glad they're partners.
Find her on X and LinkedIn at @claraveo. Check product at chatpd.ai. Follow/subscribe/like on YouTube/howiaipod, Spotify, Apple. They share fun AI stories with laughs; Polly occasionally appears uninvited.
Thanks exchanged. Lenny promotes subscribing, rating, reviewing on Apple Podcasts, Spotify, etc., and visiting lennispodcast.com for past episodes.
Keep Lenny's Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
