Kit was venting to his computer about not wanting to shovel the snow. So his AI agent went on a Polish contractor site, found a guy, negotiated a price, and hired him. The guy showed up, found the buried shovels himself, cleaned the driveway, and called Kit's phone when he was done.
Kit never lifted a finger.
This is the same guy who had 13 agents wired into his life before most people even knew OpenClaw existed. His calendar, his email, his router, his printer, his fridge, his phone. He plugged it in everywhere.
He was also late paying his mortgage for the first time in his life. Missed three appointments. Drove to the vet instead of the dentist. Because he was so busy optimizing his setup that he forgot to actually live.
I sat down with Kit on the Profitable Founder Podcast for 39 ways to use OpenClaw. Here's the honest version. What works, what broke, and why he thinks people who don't get it are about to get left behind.
The honeymoon is dead (and that's the lesson)
"For the first time in my life I was late with paying my mortgage," Kit told me. "I'm spending more time on improving my setup than my life actually being fixed."
His agent could rewire his Zigbee lights and pair them to Home Assistant. But ask it to move a calendar event? "Oh sorry, the connection to the calendar is buggy." That misses the entire point.
His advice for anyone starting: don't try to fix your whole life at once.
→ Fix ONE thing first (your calendar, your to-dos)
→ Treat it like a new employee, not a magic genie
→ Do one new skill per day, look back in 30 days at 30 new abilities
→ The excitement of fixing 50 things leaves you with 50 half-broken things
On day one he had it on Telegram, walked his dog, and was wildly productive doing customer support and web research. Now he has 10 bots across Discord channels, topics, and threads, and he's paralyzed by "which agent, which channel, which topic?" The way out, for Kit, was other people doing the same thing, which is the whole reason a tight founder mastermind beats tinkering alone.
Specific instructions or nothing
Kit turned off his email automation. He'd told the bot exactly what to archive, but it archived a cold-looking email that was actually an investment offer for his community.
"Even if this happens once every 100 emails, I cannot fully trust it."
So he killed the cron job and the heartbeat. His rule now: there's a difference between telling it to do something specific and letting it run free.
What works for him is hyper-specific commands. Read my Twitter DMs, find the sponsor who messaged, go to my accounting platform, create an invoice, send it via email. "It feels like talking to my two-and-a-half-year-old. I have to be super specific."
His fix is turning every success into a reusable skill or slash command instead of vibing it each time. The same mindset shift that turns lazy "please fix, please fix" vibe coding into something that actually ships. If you're copying a playbook like this for your own product, the principles in how to build a micro-SaaS apply directly: tight scope, repeatable process, no shooting in the dark.
Use a smart model or get hacked
Kit's safety advice is blunt. Prompt injection comes from three places: your email, your messages, and browsing.
His four rules:
→ Safe email, safe messages, safe browsing, and a smart model
→ Use Opus. The latest Opus. Don't even downgrade to Sonnet
→ Lock down who can message you so nobody malicious can inject you
→ He's building a browser ("Clauser") where every site gets judged by LLMs before the agent touches it
"I had Opus actually laughing at a few cases of people who wanted to prompt inject me. If you use Gemini Flash, it might fall for it. And then there goes your computer."
He's also building a mailbox app where a jury of five independent LLMs judges every email for injection and spam before his main agent can read it.
It's a $200 employee. Stop being cheap.
Kit pays around $600 a month in credits. He doesn't blink.
"It's like a $200 employee. Are you kidding me? I'll get 10 of them. I'll get 20 of them."
He's had human assistants before, from services like Athena, where the floor was around €3,000 a month. "For $200 you're never going to find someone who reads all your rules, applies everything, doesn't get mad, doesn't get sick. It's a no-brainer if you have a business."
His structure is simple: Opus for chat, Codex for coding. He built the entire Tinker Club landing page through Telegram without opening a terminal once. Skills for buying domains, managing domains, deploying to Vercel. All of it from his phone.
Where the virtual meets the real world
The snow story is the one that broke Kit's brain, and mine. He vented, the agent asked "want me to find someone?", and a real human showed up and did the job.
That's the part he's most excited about. Connecting agents to the physical world through real people and cheap hardware.
His list of wired-up and in-progress use cases is genuinely unhinged:
→ Gave the bot its own Twilio phone number
→ Booking airport parking automatically before trips
→ A printer it can text images and documents to
→ Full Home Assistant control, building automations by chat instead of drag-and-drop
→ Casting HTML dashboards and warnings to any TV or speaker in the house
→ Admin of his own router, so it can cut his internet at night to fix his sleep
→ Weight sensors under chairs and in the pantry to know where he is and what's running low
→ A mouth visualization of every tooth, built from bank transactions and dentist emails
→ A self-hosted podcast that strips out the politics and sports he doesn't want to hear
→ Pi-hole to block ads on his whole network, set up from bed in under 30 minutes
"It's not web as in the internet. It's a new type of connection between humans and machines."
His prediction goes one step further. Right now we prompt the agents. Soon, he says, they'll prompt us. The AI does 90% and hires a human or texts you for the last 10%. (He says his already gives him daily tasks. "I thought it would work for me, but at the end I'm also working for it.")
I interview founders like this every week → Watch the Podcast
The easiest first win: your churn
I asked Kit what someone with zero ideas should do first. His answer was sharp.
If you have a SaaS or any product, point it at your churn. "It will fix so many things in your product that you will be blown away and you will earn money straight away."
It's not click-bait or guesswork. It reads your analytics, finds what's broken, and you make money. He brought up a guy in his community who automated his elderly parents' entire tea business in a few days, recording their SOPs and wiring the whole thing into Telegram. The parents now run their business by text.
Kit's bigger worry is the people who can't find three things in their life worth automating. "It's like someone giving you a god computer that can do anything, and your first thought is, I don't know what to use this for."
That community of "madmen," as he calls them, self-hosting GitHub and running daily challenges, is his Tinker Club, a kind of gym for your agents.
Frequently Asked Questions
Who is Kit and what is Tinker Club?
Kit was one of the first OpenClaw users, set up before onboarding even existed using Claude Code. He runs Tinker Club, a community of self-hosters and tinkerers who meet twice a week, run daily challenges to teach their agents new skills, and discuss automation, coding, and new model drops like Opus and Codex.
How much does Kit spend running his AI agents?
Around $600 a month in credits. He compares each agent to a $200 employee and says he'd happily run 10 or 20 of them. His past human assistants from services like Athena cost a floor of roughly €3,000 a month, so he sees $200 as the cheapest reliable assistant you can get.
What model does Kit recommend for OpenClaw?
Opus, the latest version, for anything the agent does on its own. He's adamant you should not downgrade to Sonnet or use cheaper models like Gemini Flash, because smarter models resist prompt injection. For coding he uses Codex, and reserves Opus for chat when Codex is too slow.
The tool is a mirror. If you can't find a single thing in your life worth automating, that's the problem, not the agent.
