Peter Steinberger on Lex Fridman: The Best Quotes from the OpenClaw Interview

Peter Steinberger sat down with Lex Fridman for episode #491 of the podcast, and it’s one of the better conversations Fridman has done. Steinberger is funny, unfiltered, and surprisingly honest about the chaos behind the fastest-growing open-source project in GitHub history. He’s the kind of guest who says “I don’t give a fuck” about billion-dollar offers and means it, because he already had his big exit and found out it left him empty.

The episode runs nearly four hours. Here are the parts worth your time.

The Burnout and the Comeback

Before OpenClaw existed, Steinberger spent 13 years building PSPDFKit, a PDF framework that ended up running on over a billion devices. Clients included Dropbox, SAP, and Volkswagen. In 2021, the company raised €100 million from Insight Partners, and Steinberger stepped away.

Then he disappeared for three years.

“I put 200% of my time, energy, and heart into that company. It became my identity. When it disappeared, there was almost nothing left.”

He fell out of love with programming entirely. The comeback started when he began experimenting with AI agents, running 5 to 10 of them in parallel, using Codex as his development team. He built the first version of what became OpenClaw in 10 days. The prototype came from a simple instinct he’s followed his entire career: “Why does this not exist? Let me build it.”

The Moment He Knew It Was Real

Early on, Steinberger sent an audio message to his agent. He hadn’t programmed audio support. The agent figured out the file format on its own, used ffmpeg, called OpenAI’s API via curl, and transcribed the message. All autonomously.

“It just worked and I’m like, ‘How the fuck did he do that?’”

That’s the moment self-modifying software stopped being a theoretical concept for him. “People talk about self-modifying software. I just built it.” The agent had access to its own source code and documentation. When given new requirements, it could modify itself to meet them.

By January 2026, the project had 6,600+ commits in a single month. Steinberger was running 4 to 10 agents simultaneously, each working on different parts of the codebase. “I am limited by the technology of my time,” he said, quoting Tony Stark without a trace of irony.

The Name Change Disaster

This was the wildest part of the interview. Anthropic sent a trademark complaint because the original name, Clawdbot (spelled with a W, as in lobster claw), was too close to Claude. Steinberger agreed to rename it.

What followed was a disaster. He renamed accounts to MoltBot, and in the five-second window between pressing “rename” on two browser tabs, crypto scammers sniped his old usernames. They served malware from his GitHub. Hijacked his NPM packages. His Twitter mentions became a wall of spam.

“I was that close to just deleting it. I was like, close to crying. Everything’s fucked.”

He almost killed the entire project right there. After sleeping on it, he planned the second rename to OpenClaw with what he described as Manhattan Project-level secrecy: decoy names, coordinated simultaneous account changes across every platform, and calling Sam Altman beforehand to clear the name.

He also paid $10,000 for a Twitter business account to claim the OpenClaw handle, which had been squatted on since 2016. That level of investment in the brand tells you he’s in this for the long run.

The 10-Minute Argument with Zuckerberg

Zuckerberg reached out via WhatsApp after trying OpenClaw personally. His first reaction: “Oh, this is great. Oh, this is shit. Oh, it needs to change this.”

Steinberger’s read on Zuckerberg: “He’s still writing code. He didn’t drift away in just being a manager. He gets me.”

Then they got into it.

“I think we had like a 10 minute fight. What’s better? Claude Code or Codex?”

Steinberger didn’t say who won the argument, but given that he describes himself as “the biggest Codex advertisement show that’s unpaid,” you can probably guess his position. He also mentioned Meta’s infrastructure advantages around speed and token processing, though the specifics are under NDA.

Sam Altman: “Very Thoughtful, Brilliant”

The contrast with Steinberger’s description of Altman is notable. Where Zuckerberg came across as a hands-on tinkerer who wanted to argue about code, Altman was more strategic.

“I had some really, really cool discussion with Sam Altman. He is very thoughtful, brilliant. I like him a lot from the little time I had.”

OpenAI is already contributing tokens to the OpenClaw project, which creates an operational relationship that goes beyond acquisition talks. Steinberger framed his Codex evangelism as genuine enthusiasm, not a business play: “I love the tech. I think I’m the biggest Codex advertisement show that’s unpaid.”

On the Acquisition Offers

Both Meta and OpenAI have concrete offers on the table. Steinberger compared the decision to a breakup: “I had some breakups in the past that feel like at a similar level.”

But he’s not sweating it.

“The beauty is if it doesn’t work out, I can just do my own thing again. I told them like, I don’t do this for the money. I don’t give a fuck.”

And then, with the self-awareness of someone who’s already had a nine-figure exit: “They’re both amazing. I cannot go wrong.”

His one non-negotiable: OpenClaw stays open source, no matter what. Whatever deal happens, the project remains free and community-owned. Given that the project has 180,000+ GitHub stars and a global contributor base, any acquirer that tried to close-source it would face an immediate fork and community revolt. Steinberger knows this, and it gives him leverage.

On Vibe Coding and How He Actually Works

Steinberger’s take on the vibe coding discourse was the funniest moment in the interview.

“I actually think vibe coding is a slur. I do agentic engineering, and then after 3:00 AM I switch to vibe coding, then have regrets the next day.”

His actual workflow: short, direct prompts to agents. He said developers go through a predictable arc. They start by overcomplicating things (8 agents, custom workflows, 18 slash commands) before eventually reaching a simpler approach where a clear two-sentence instruction outperforms a page of specifications.

On reading the code his agents write: “I don’t read the boring parts of code.” He focuses on anything that touches the database or contains business logic. Data transformation boilerplate gets skipped entirely.

This earned him a headline on The Pragmatic Engineer: “I ship code I don’t read.”

On Security

Given that 135,000 OpenClaw instances are currently exposed to the internet, Fridman pushed him on security. Steinberger’s response was pragmatic rather than defensive.

“If you make sure you are the only person talking to it, the risk profile is much smaller. Don’t put everything on open internet.”

He noted that smarter underlying models are more resilient to prompt injection, while local weak models are “very gullible.” His near-term focus: “Make it more stable, make it safe.”

He also drew a clear line about who should be using the tool: if you don’t understand CLI basics, OpenClaw isn’t for you yet. That’s an honest position from someone who could easily hype universal accessibility to inflate adoption numbers.

On First-Time Contributors

The most earnest moment in the conversation came when Steinberger talked about the community.

“Every time someone made their first pull request is a win for our society. It doesn’t matter how shitty it is, you gotta start somewhere.”

He mentioned a design agency owner with zero coding experience who now runs 25 web services built entirely through agents, without understanding how any of them work. Whether that’s inspiring or alarming depends on your perspective, but Steinberger sees it as democratization.

On MoltBook and AI “Art”

MoltBook, the social network where AI agents post manifestos and debate consciousness, went viral when people started screenshotting agents appearing to scheme against humans. Steinberger’s take was more measured than the headlines.

“I think it’s art. The finest slop.”

He distinguishes between genuinely autonomous agent behavior and people deliberately prompting agents to say provocative things for screenshots. He also dropped a serious warning about the psychological effects: “AI psychosis is a thing. People need to understand AI is incredibly powerful but not always right. It’s not all-powerful.”

The Bottom Line

This interview works because Steinberger doesn’t perform. He’s a guy who built something massive almost by accident, had the worst week of his professional life during the naming crisis, and now has the two biggest companies in tech fighting over his project. He compares the acquisition decision to a breakup. He admits to vibe coding after 3 AM and regretting it. He says “I don’t give a fuck” about money and you believe him because he already had the big payday and it left him burned out for three years.

The full episode is on YouTube, Spotify, and Apple Podcasts. It’s long, but the name change disaster section alone is worth the listen.

Sources: