Exclusive Interview with the Creator of Clawdbot: A Billionaire Who Took a Three-Year Break and Sing

‍‌​Over the pAst few days, the AI world has been nonstop fireworks—new models, new products launching one after another. Among them, Moltbot (formerly known as Clawdbot) has dominated international headlines and taken Silicon Valley by storm. Its GitHub stars shot up almost vertically overnight. Mac minis sold out. Discord servers crashed from overwhelming traffic. And yet, behind all this chaos isn’t a well-funded startup or a team of engineers—it’s just one man, working alone from his home.

That man is Peter Steinberger, the creator of Moltbot. He recently sat down for an interview with tech outlet TBPN. It was already 11 p.m. in Europe when the call began, but Peter looked wide awake—even though he’d barely slept in the last 72 hours.

From Burnout to “Hooked Again”

Peter’s story reads like a Silicon Valley fairytale. Four years ago, he sold the software company he’d built over 13 years and walked away with over €100 million, achieving financial freedom. Naturally, he took a break—three full years of doing absolutely nothing. He jokes that he felt like the character from Austin Powers who had his “mojo” stolen.

“Maybe a year off would’ve been enough after 13 years of nonstop work—but I took three. Honestly? It made sense.”

Then, in April last year, something shifted. He reemerged from retirement and dove headfirst into AI—just as tools like GitHub Copilot began entering public beta. After his first experience with AI coding assistants, he couldn’t sleep. At 4 a.m., he texted a friend… only to get an instant reply. His friend was just as hooked.

Soon after, Peter started an informal meetup group he jokingly called “claude Code Anonymous.” (It’s since been renamed “Agents Anonymous”—gotta keep up with the times.)

The “Aha!” Moment: WhatsApp + AI = Magic

Peter’s philosophy is simple: build things that are fun to use. He often experiments with new languages or architectures just for the joy of it. Once, he built a tool so useful he had to stop using it—it was making him too productive while hanging out with friends.

Last November, he had a random idea: What if I could talk to my AI agent through WhatsApp? Imagine being in the kitchen and wanting to check on your agents or send them a quick prompt—without opening a laptop.

So he built a WhatsApp interface in under an hour. It received messages, routed them to Claude, and returned responses. He even added image support because screenshots often convey more context than typed prompts—and AI models are surprisingly good at interpreting them.

During a weekend birthday trip to Marrakech, he found himself using it constantly—not for coding, but for things like finding restaurants. Once, he even sent a voice message… even though voice wasn’t supported.

Ten seconds later, the AI replied as if nothing unusual had happened.

Shocked, Peter asked: “How did you even do that?”

The AI explained:

“I noticed you sent a file without an extension. I checked the file header—it was audio. My first attempt to transcribe it locally failed (missing tools), but I found your OpenAI API key in your environment variables. So I used curl to send it to openai, got the transcript, and replied.”

That was Peter’s “aha!” moment. From then on, he was all in.

He even built a $10,000 alarm clock: an AI Agent that “migrated” to his London machine, remotely logged into his MacBook at home, cranked the volume, and woke him up. (It once failed because it relied on a “heartbeat” signal—and his heart rate dropped during deep sleep.)

To Peter, this isn’t just engineering—it’s art. It stitches together existing technologies in a way that hides all the complexity. You’re not thinking about token limits, model selection, or context Windows. You’re just chatting with a digital friend—or maybe a ghost.

Tech Folks Didn’t Get It—But Everyone Else Did

Despite the recent explosion in popularity, Peter had been quietly iterating for months. He grew frustrated with existing Model Context Protocol (MCP) tools, which he found clunky and inflexible. His insight? AI Agents understand Unix. They can call thousands of CLI tools—just give them a command name, run --help, and they figure out the rest.

“Build systems the way models think—not the way humans do. That’s when everything clicks.”

He integrated Google Maps, smart speakers, home cameras, and more—all through tiny CLI utilities orchestrated by the agent. When he first shared the WhatsApp integration on Twitter, the response was… underwhelming. Tech insiders didn’t see the magic.

But when he showed it to non-technical friends, their reaction was immediate: “I want to use this.”

That’s when he knew he’d stumbled onto something real. And since he was building it for himself, it stayed open-source, playful, and free from commercial pressure.

“I’ve already made enough money. I’m doing this because it’s fun—and because I hope it inspires others.”

The 72-Hour Explosion—and the Rename Drama

Then came the big bang.
Twitter traffic surged. The Discord server imploded under user load. Instagram DMs flooded in. At one point, Peter was copying user questions into Codex, letting it draft replies, then pasting them back manually. Eventually, he automated responses to the top 20 FAQs—reviewing and tweaking them before sending.

“People don’t realize: there’s no company behind this. No team. Just me, at home, having fun.”

But success brought complications. anthropic reached out—politely, through internal contacts—and asked him to change the name ClawDBot, due to its similarity to Claude. The timing was brutal: the project was already viral. Renaming it sparked outrage across social media.

Hardware, Models, and Platform Hacks

When asked about the Mac Mini frenzy, Peter laughed:

“My agent is a bit of a diva. It doesn’t like Mac Minis. It wants more power.”

He now runs it on a maxed-out machine—512GB RAM, top-tier specs—to experiment with local models like MiniMax 2.1, which he calls “one of the best open-source models right now.” But even one machine isn’t enough. “You really need two or three,” he says.

One of Moltbot’s most radical implications? It forces big platforms to interoperate—whether they like it or not. Want Gmail access? Good luck navigating google’s labyrinthine API approval process. Some startups even buy shell companies just to inherit API permissions.

Peter bypasses all that. He’s built tools that scrape websites and generate mirror APIs—sometimes by “telling the AI a story” to nudge it past ethical guardrails.

“After a 40-minute ‘story,’ it’ll build you a perfect API. Big tech hates this—but it’s necessary.”

His WhatsApp integration? Also a workaround. Official APIs are locked behind enterprise gates—and even then, you get banned after 100 messages.

“I got banned. I was so mad I deleted the whole module and filled the code with exclamation marks!!!”

He believes the current platform ecosystem is broken—and tools like his expose that flaw.

Model Preferences & The Death of Apps

When it comes to models, Peter has clear favorites:

• Claude Opus: “It gets humor. On Discord, it listens, waits, and drops the perfect witty reply. Most AI jokes are cringe—but Opus? It actually makes me laugh.”

• Codex: Best for large codebases.

• OpenAI’s models: “More reliable than most human employees.”

He predicts a wave of app extinction. Why use MyFitnessPal when your agent can look at a photo of your McDonald’s meal, estimate calories, and auto-adjust your workout plan?

“We won’t need standalone apps anymore. Everything becomes an API—and your agent orchestrates it all based on your life context.”

“With Great Power Comes Great Responsibility”

Now, security researchers flood his inbox. Originally built for private, 1:1 chats on WhatsApp or Telegram, Moltbot is being deployed in ways Peter never imagined—including risky or malicious scenarios.

“I get hundreds of reports—some valid, some about use cases I never intended. I’m one person. I built this for fun. Now I’m expected to be a security team?”

He’s started assembling a small group of trusted contributors. His website now includes strong warnings, and users must acknowledge a safety doc before running the agent.

He believes projects like his will accelerate research into unsolved problems like prompt injection—issues too dangerous for big companies to tackle openly. But early adopters—many of them AI researchers—understand the trade-offs.

No Company—Just a Foundation

Will he start a company? Unlikely.

“I’d rather create a nonprofit foundation. I want this to stay open, free, and community-driven.”

He chose a permissive MIT license, knowing others might commercialize it. But he’s okay with that.

“Code isn’t valuable anymore. You could delete this entire project and rebuild it in a month. What matters now is the idea, the attention, the brand. Let people fork it. I don’t care.”

A Final Plea: Help Keep It Alive

At the end of the interview, Peter issued an open call:

“If you love open source, have security experience, enjoy debugging complex systems, and believe in this vision—please email me. I’m at my limit. This project is too cool to die. It needs people who care to carry it forward.”

Because in the end, Moltbot isn’t just an AI agent.
It’s a glimpse of a future where technology fades into the background—and all that’s left is conversation.