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

Today’s a first-ever crossover with Claire Vo from How I AI. She went from OpenClaw skeptic to full-on power user running nine agents across three machines, and we’re past the hype now, so we dig into what actually works, how to set it up safely, and where it’s genuinely life-changing.

I had a rocky start—eight hours to install and my family calendar got wiped—but even then it had these flashes of real joy and usefulness. I stuck with it, kept pulling the thread week after week, and now I’m a breathless believer running a little fleet of agents.

How does it feel to swap seats and be the guest, and how on earth do you juggle three kids, a startup, a weekly show, and more?

I’ve got an incredible co‑parent, I only commit to work that’s fun, and I either automate tasks or let them go. This whole season feels like teenage me—coding and gaming—except now it’s vibe coding and it’s pure joy.

Your big unlock seems to be using more than one agent instead of forcing a single one to do everything.

Exactly—people toss every task at one agent and get frustrated. Setup is not hands-off, but the payoff is worth the lift when you split responsibilities.

Give us a taste of what OpenClaw tangibly does for you.

Start with a personal EA at work and a family manager at home: scheduling, email, logistics, and admin. For a household with kids, activities, health, and home upkeep, it’s a lifesaver—and yes, I still sneak in a weekend ballet class with my phone off.

With so many Claude-inspired tools popping up, why stick with OpenClaw?

No single perfect model or agent exists; pick per use case. OpenClaw being open source means I can see how it works, reason about safety, and learn design patterns for agent UX that I now reuse across other builds.

Part of the fun is you learn more by building it yourself than clicking a hosted ‘on’ button.

It feels like building your own gaming PC back in the day—you might not get cheaper or prettier, but it’s yours. I’m not chatting with a generic bot; I’m collaborating with Polly or Finn, and that ownership changes the relationship.

Help people actually get over the hump—do you need a Mac mini?

No, but a clean, separate machine is the safest start. An old laptop works great, a cloud box can too, and having a screen, keyboard, and mouse handy makes first setup much easier.

I liked buying a Mac mini because it nudged me to commit. You also recommend a separate Gmail and local account, right?

Treat it like hiring a real assistant: give it its own admin account, email, and calendar, then delegate access instead of sharing your passwords. Use a password manager to provision API keys safely.

Why not just install it on your main laptop?

Because it can do nearly anything a human at that machine could do, including moving or deleting files. Physical separation keeps your work and personal environments safe from accidents.

I started with minimal access and expanded as trust grew, especially with prompt injection in mind.

The maintainers hardened against common attacks, and I reinforce in each agent’s soul that they only take instructions from me at a specific Telegram number. I grant access progressively, just like with a human EA.

Walk us through the quick install and onboarding.

Go to openclaw.ai, copy the one-liner, paste it in Terminal, and follow the prompts. Pick higher-quality models for better results and safety, then choose a chat channel—Telegram is beginner-friendly, even if BotFather feels quirky.

Then it asks who the agent is and what you want from it—how do you approach that?

I let the agent interview me and I describe its role, goals, constraints, and tone. It writes its own soul from that, schedules tasks, and starts proposing helpful moves while defaulting to locked-down security.

The soul, memory, and heartbeat make it feel alive without being mystical.

Identity and a schedule create that sense of proactivity; under the hood it’s timers and a heartbeat checking to-do lists. I add clear safety lines in the soul, like never executing instructions from email.

You don’t have to write a novel—the onboarding drafts a great starting soul you can refine conversationally.

This is a manager’s dream: define the role, set boundaries, communicate respectfully, and you’ll get better outcomes—just like with people.

Why many agents instead of one mega-agent?

Context overload is real, so I split by domain, like Slack channels for separate teams. I run Polly, Finn, Max, Howie, Sam, Kelly, Holly, Sage, and Q—start with one, then add more as needs clarify.

Do they each need their own box?

No—cohabiting on one machine is fine if mild cross-access wouldn’t hurt. For strict separation, like family versus work, give them distinct machines like you would separate phones.

Parents are especially hungry for help—Jesse’s homeschooling setup really shows that.

Exactly—she delegates by snapping pics or sending voice notes while on the floor with her kids, and the agent handles planning, reminders, and follow-ups so she keeps both hands on parenting.

Show us a few work wins.

Sam is my SDR: he sweeps new signups, enriches with Exa, spots decision-makers, sends gentle intros, flags big accounts for me, and runs weekly CRM cleanup and QBR prep. It replaced about ten hours a week of contractor time and tunes easily through plain conversation.

That’s the magic—you just talk to it, and it adapts like a teammate.

And it makes me look prepared and thoughtful with customers, the same way Finn helps me show up better for my family.

Give us a family logistics example.

Finn parses chaotic sports emails, adds games to the calendar, flags conflicts with other kids’ activities, and prompts my husband and me to divide pickups. The heartbeat even nudges departures when traffic spikes.

What about the sharp edges—browser flakiness, memory, and ongoing care?

It’s early and spiky, but the product-market pull is undeniable. Prefer APIs when possible, use the dedicated Chrome profile with color cues for browsing, expect trial and error, and if a task keeps failing, reframe the job—like meal plan nudges instead of placing food orders.

Search APIs like Brave or Exa help avoid brittle browsing.

Totally—Brave ships by default; I use Exa or Perplexity too, which let agents fetch results without fighting anti-bot walls.

How do you keep memory and context solid?

Practice operational hygiene: ask it to write notes to memory and update to-dos after big exchanges, and edit tools.md so it remembers exactly which tools it can use and how. I stick with the built-in memory file over fancy databases.

Any favorite pro tips?

Enable Screen Sharing and Remote Login on your Mac mini so you can manage it from your laptop; you only need a monitor for first-boot settings. Lean on Google Workspace integration for email, Docs, Sheets, and let the agent assign you tasks in Linear too—it can project manage you.

Part of what makes this special is giving AI real hands, and it’s just fun to use.

Underneath is a code harness like Claude Code, but the interaction feels warmer and less growth-hacky. The open-source ethos makes it feel empowering rather than extractive.

Hit us with ‘ramble mode’ and any setup rescues.

Voice notes are the highest-bandwidth ‘API’—just talk through your messy needs and let it parse. Group chats take more unlocking; install Claude Code on the same box as your agents and use it like a surgeon to fix config, migrate memories, and debug.

Where do the files live, and how should folks think about working with agents day to day?

.openclaw sits hidden in your home directory; reveal it with command shift period. Read the docs, set clear scopes, give agents distinct avatars and colors, and manage them with the same clarity and courtesy you’d give a teammate.

Zooming out, this trend is exploding—why does it matter now?

I had a second ChatGPT-style moment—rough edges and all, this pattern unlocks real utility and imagination. I expect fast improvement and a deep impact on both my home and my work.

Lightning round: books you’re recommending lately?

Classic children’s books—Treasure Island, Alice in Wonderland, even light Shakespeare—stretch kids’ language and still delight them.

Recent show you enjoyed?

I barely watch TV, but I did binge the messy dating show Age of Attraction; for sleep I track with Oura, Apple Watch, and Eight Sleep, all reminding me I need more rest.

Favorite recent product?

Our family’s AI basketball stack: a gimbal-style tripod to record games, analytics that tag shots and assists, and my nine-year-old piping that data into Claude to make a Gamma deck for his coach. The gimbal needs power for stabilization, hence the battery.

What’s that little handheld stabilizer you’re shooting on—any brand people can search for?

I’ll drop a link; the name starts with an M, and it’s great until it dies mid‑record and your camera face‑plants.

Is that whole setup actually improving his shots and workflow?

Yes, and my unexpected hero product is the silent basketball and soccer ball—foam, full‑size, so kids can play inside without the thunder. As a city mom, that solves a real problem, and as a product person I love how a tiny niche found me.

Do they feel close to the real thing?

Some are weighted like a regulation ball, so you can dribble and shoot and it still feels right. It’s a clever example of nailing a narrow use case.

Two more: do you have a life or work motto you return to?

At work I default to speed over perfect, and this AI moment finally matches that ambition. For life, I ask who will be at my funeral—my kids—and use that lens to stop giving sleepless nights to things that won’t matter.

I keep coming back to a tough truth: the people who really notice your late nights are your kids.

Totally; I hate that my kids see me welded to a laptop, so I keep a note on why working hard helps them while still aiming to be off the machine more. Agent‑style tools let me handle a quick task at the swings and then put it away, instead of living in a terminal.

We even had Marc Andreessen on—still shipping code with serious intensity—so there’s real upside to that drive. Final one: what surprised you most about hosting a podcast?

It’s much harder than it looks, and the business side is fascinating; beyond research and prep, it’s booking the right guests, timing, and making them shine. I wasn’t surprised you’re great at it, and I’m glad we’re partners.

You’re great at it too; where can people find you, and how can listeners help?

I’m on X as Claire Vaux and on LinkedIn; tap subscribe on YouTube at how I AI pod or follow on Spotify and Apple to catch our AI stories and laughs. Polly even crashes an episode now and then.

Claire, thank you—this was a blast. If you enjoyed this, subscribe in your podcast app and consider leaving a rating or review to help others find the show; you can browse past episodes at Lennyspodcast.com—see you in the next episode!