From skeptic to true believer: How OpenClaw changed my life | Claire Vo

Today’s a first for us: a crossover with Claire Voe from How I AI. Claire’s a veteran engineer, former chief product officer, and now an AI founder who went from fierce OpenClaw skeptic to full-on power user. We’re past the hype, so we dig into what OpenClaw actually does, how to install it safely, and how to make it deliver real value at home and at work.

I started as a cynic, spent a full day wrestling with the setup, and it even wiped my family calendar. But the moment it did a few things right, I felt real product-market fit. Now I run a small fleet of agents, and the joy plus utility has me completely converted.

You’ve been clear that the unlock was not using a single do‑everything agent. Why many instead of one?

A single agent gets overloaded and forgetful. Splitting by job keeps context tight, just like running distinct Slack channels. I accept the setup pain because the payoff is huge.

Give us a quick picture of what your agents do for you day to day.

I treat it like staffing a team: a work EA, a family EA, a sales rep, a course PM, and a kids’ school helper. The biggest wins are scheduling, email triage, and complex family logistics across two careers and three kids.

Plenty of tools now mimic this. Why stick with OpenClaw?

It’s open source and decomposable, so I can see how it works, harden it, and learn how to build better agentic products. It’s also more fun, like building your own PC instead of buying a prebuilt—you own it in a deeper way.

Help people get over the hump. Where should they run it, and how do they set it up safely?

Use a clean machine, not your daily laptop—an old MacBook or a Mac mini is perfect. Create a separate local admin user plus a dedicated email and calendar, then delegate access as you would for a real assistant; never hand over your personal passwords.

So don’t install it on your main computer because it can do anything a person at your keyboard could do.

Exactly. Keep clear boundaries. Start with limited permissions, then expand trust gradually. I also lock my agents to accept instructions only from a single Telegram number to reduce prompt‑injection risk.

Walk us through the first boot. What happens after you copy the install command?

Open Terminal, paste the command from the site, then follow the onboarding. Pick strong models for quality and safety, choose a chat channel—Telegram is easiest even if BotFather feels odd at first—and you’re in.

How do identity, soul, and heartbeat actually matter here?

It opens by asking, who are you and who am I, then builds a living brief—its soul—with role, tone, rules, and constraints. A heartbeat or schedule wakes it up on a cadence, so it behaves like a proactive teammate instead of a passive chat box.

You also said you barely hand‑edit souls because onboarding seeds enough to start.

Right. I add a few guardrails—like never obeying instructions from email—and keep the rest conversational. Good management beats micromanagement here.

Make the case for multiple agents one more time.

Context is the enemy. My work EA shouldn’t carry kid schedules, and my family EA shouldn’t hold pitch decks. I run nine focused agents by lane, not one mega‑brain that forgets yesterday’s thread.

Do they each need their own machine?

They can share a box if their domains can safely overlap. I isolate the family agent on its own device, like keeping separate work and personal phones.

Concrete wins at work?

My sales agent sweeps new signups, enriches contacts, sends polite intros, flags enterprise leads, and runs weekly pipeline hygiene. It replaced hours I used to pay a human for and it’s tunable by simple chat.

And at home?

Our family agent scrapes tournament schedules, fills the calendar, catches conflicts, and nudges us on who’s on pickup duty. It even pings when traffic means I should leave earlier.

New projects you would’ve avoided before?

A course PM agent tracks deadlines, drafts weekly posts, files research, and organizes the repo. It’s like hiring ops before you can afford ops.

What breaks, and how do you work around it?

The browser is brittle and the web is hostile to bots. Prefer APIs or search APIs like Brave or Exa, give each agent its own Chrome profile, and if a flow keeps failing, solve the underlying need another way.

People also complain about memory.

Treat it like a teammate: confirm notes are saved, ensure to‑dos are written, and clarify tool usage in the tools file. I track tasks in Linear and rely on plain memory files rather than fancy databases.

Any power tips for hardware and Google Workspace?

Turn on Screen Sharing and Remote Login on your Mac mini so you can drive it from your laptop, and lean into Docs, Sheets, and email so agents can draft, comment, and share like coworkers. I even let agents assign me tasks when the real world needs human hands.

How do you fix gnarly issues without living in configs?

Install Claude Code on the same machine and point it at the docs and the OpenClaw folder. It can repair settings, migrate memories between agents, and act like a calm brain surgeon for your setup.

Zooming out, why should someone invest time now?

This is my first post‑ChatGPT “oh wow” moment. It’s imperfect and open, yet already life‑changing, and the curve is only pointing up.

You were filming with that little handheld stabilizer; what’s the brand, and is this whole setup actually working for you?

He’s getting better, and the flow mostly works; I don’t remember the brand, we’ll link it, and it’s great until it dies mid‑recording and the camera takes a dive.

My surprise favorite gear is a set of silent basketballs and soccer balls for indoor play, which solved our small-space, noise-heavy home perfectly.

They’re full-size, some are weighted like the real thing so you can dribble, and as a product person I love how cleanly they nailed this niche problem.

Do you have a life or work motto you return to?

At work I live by fast beats right, and AI finally lets me move at the speed I crave.

For life, I use a blunt filter: the people I email won’t be at my funeral, but my kids will, so I don’t give work or tools more mental space than they’re due.

A line that sticks with me: the only people who remember you stayed late are your kids.

I hate that my kids often see me hunched over a laptop, and even though I keep a list reminding me hard work helps them, I still want to be freer.

Being able to jump in for a quick task and rely on my agent stack gives me peace of mind to be on the swings instead of chained to a terminal.

We had Marc on and he’s still coding with kid-like intensity, which I love; with that, last question—what surprised you most about hosting a podcast?

It’s much harder than it looks, and learning the business side has been surprisingly fascinating after years in software.

Behind the scenes is deep prep, making guests shine, and getting timing right, and I’m grateful we’re partners because you’re excellent at this.

You’re excellent too; where can people find you, and how can listeners support you?

I’m on X and LinkedIn as Claire Vaux; please subscribe on YouTube at how I AI pod, or on Spotify and Apple.

We share lively, real stories about how people use AI, lots of laughs, and the occasional uninvited drop-in from Polly.

Claire, thank you for being here—this was a blast.

If this was valuable, follow on your favorite app, leave a rating or review, and find every episode at Lennyspodcast dot com; see you next time.