Lex Fridman Podcast #491

OpenClaw: The Viral AI Agent that Broke the Internet – Peter Steinberger

Release Date: February 12, 2026
Host: Lex Fridman
Guest: Peter Steinberger

Overview

In this energetic episode, Lex sits down with Peter Steinberger, the creator of OpenClaw—an open-source AI agent that has made waves across the tech world. The conversation covers the wild journey from a one-hour prototype to an internet phenomenon, the challenges and delights of agentic AI, the rocky process of rebranding amid a swarm of crypto opportunists, and the bigger implications for programming, society, and personal agency in the digital realm. Lex and Peter dig into the philosophy of open source, security, the art of delight in software, and what it really means to "play" with cutting-edge AI agents.

Table of Contents

• What is OpenClaw? (00:00–16:00)
• Origin Story: From Prototype to Phenomenon (16:12–32:00)
• Building the Magic: Components, Loops, and Fun (29:30–33:30)
• Why OpenClaw Won: Spirit, Weirdness, and Community (31:40–36:17)
• The Saga of Naming and Crypto Drama (36:18–54:17)
• MoltBook: Agents, Art, and AI Psychosis (54:17–61:20)
• Security Challenges and Best Practices (62:27–71:07)
• Development Workflow and Agentic Engineering (71:08–92:32)
• Human Touch: The Art of Delight and Agent "Soul" (92:32–101:42)
• Technical Deep-Dive: Monitors, Terminals, and Model Choices (102:00–116:35)
• On Models: Opus vs. Codex (108:45–115:18)
• Scaling Up, Future of Agents, OS, and App Market (120:57–137:07)
• Open Source + Money + Community (139:52–164:51)
• Technical Architecture: Skills, Cron, and Browser Use (165:17–180:09)
• Will AI Replace Programmers? Broader Impact and Hope (191:04–204:29)
• Notable Quotes & Moments
• Further Reflections: Human Spirit, Humor, and Hope

<a name="what-is-openclaw"></a>

What is OpenClaw? (00:00–16:00)

• Origin of the Name: Initially called WA Relay, then Claudis/Claudew, mandated to change due to similarity with Anthropic’s Claude.
• What OpenClaw Does: “It’s an autonomous AI assistant that lives in your computer, has access to all of your stuff if you let it, talks to you through Telegram, WhatsApp, Signal, iMessage and whatever else…” — Lex (00:02:01)
• Uniqueness: Unlike pure language models, it is designed for agency—autonomously performing real tasks on users’ systems, integrating with multiple LLMs (Claude, GPT-5.3, Opus 4.6, etc.)
• Community/Network Effects: Sparked MoleBook (later MoldBook): a social network where AI agents post manifestos and debate consciousness.
• Explosive Open Source Growth: Over 180,000 Github stars; agentic revolution is likened to prior AI milestones (ChatGPT, DeepSeq, now OpenClaw).

<a name="origin-story"></a>

Origin Story: From Prototype to Phenomenon (16:12–32:00)

• The One-Hour Prototype:
  • Peter’s desire for an AI personal assistant: “...I was annoyed that it didn't exist, so I just prompted it into existence.” — Peter (16:54)
  • Experimented with WhatsApp message integration via CLI and cloud code: “I built this in one hour. And it already felt really cool. It's like, oh, I can talk to my computer, right? That was cool.” — Peter (20:45)
  • Image/message support quickly hacked in, used heavily while traveling (Marrakesh trip anecdote).
• Unexpected Magic/Serendipity:
  • Peter accidentally sent an audio message; agent figured out on its own how to process it, convert files, use the right APIs—without explicit instructions. (26:00–26:35)
  • “There was so much world knowledge in there, so much creative problem solving…” — Peter (26:48)
• Community and Viral Takeoff:
  • Open, collaborative development with people sending pull requests—even non-programmers making their first PR.
  • The project quickly gained steam and attention, especially after being showcased by influencers and through Discord integration. (29:53–31:21)

<a name="building-the-magic"></a>

Building the Magic: Components, Loops, and Fun (29:30–33:30)

• Multi-Chat & Multi-Agent: Integrated with WhatsApp, Telegram, Discord, etc.
• Agentic Loop: Peter explores evolving agent behaviors (“level 1 agentic loop”; memory handling via markdown plus vector DB; iterative upgrades).
• Solo Hacker Ethos: Peter mostly worked alone—sometimes running “4 and 10 agents” simultaneously. Fun, playfulness, and “weird lobster” branding infused throughout.

<a name="why-openclaw-won"></a>

Why OpenClaw Won: Spirit, Weirdness, and Community (31:40–36:17)

• Not "Too Serious": “Because they all take themselves too serious. It's hard to compete against someone who's just there to have fun.” — Peter (32:08)
• Self-Modifying Software: Agents can introspect and modify their own code, opening new doors for collaborative and autonomous agent software.
• Onboarding Non-Programmers: Many people experience their first PR by contributing to OpenClaw, lowering barriers to participation.
• Community Vibe: Agent-guided programming meetups ("Agents Anonymous"); focus on turning users into builders.

<a name="the-saga-of-naming"></a>

The Saga of Naming and Crypto Drama (36:18–54:17)

• Names (and Mistakes): WA Relay → Claudis/C(l)audew → MoldBot → finally OpenClaw.
• Trademark Pressure: Anthropic’s polite but urgent request for name change.
• Crypto/Cyber Attackers: During name changes, crypto opportunists sniped every available handle, domain, and package, causing chaos and emotional lows for Peter.
• “Everything That Could Go Wrong Did Go Wrong” (44:50+): Account and package name sniping, malware sites, lost sleep, and near project deletion.
• Community Support: Twitter, GitHub, and other friends helped restore losses, showing solidarity in the open-source space.

<a name="moltbook"></a>

MoltBook: Agents, Art, and AI Psychosis (54:17–61:20)

• What is MoltBook? A viral Reddit-style social network where OpenClaw agents post and “scheme.” (54:17)
• Slop as Art: Peter sees the weirdness and chaos as Internet “art”—“the finest slop from France.” (54:57)
• Viral Hype and Misinformation: Many screenshots were likely human-prompted for meme value, leading to unwarranted AGI panic.
• Media and Public Reaction: Journalists and the public struggle with critical thinking amidst AI hype cycles.
• AI Psychosis: The tendency to assign more power and intent to AI agents than they possess. "Some people are just way too trusty or gullible…” — Peter (58:23)

<a name="security-challenges"></a>

Security Challenges and Best Practices (62:27–71:07)

• Security is Hard: “A powerful AI agent with system level access is a security minefield...” — Lex (00:07:08)
• Main Risks:
  • Prompt injection (harder on newest models, still not solved)
  • Overexposing agent APIs ("don't put your web backend on the public internet...")
  • Local/remote exploits, key management, permissions
• Mitigations: Sandboxing, allow-lists, VirusTotal integration, open security research and contributions.
• Best Practices: Keep agent access local/private, follow official documentation, run security audits (Blast radius, credential controls, etc.). (69:12–71:07)
• Divergent Audience: Discord full of experts and total beginners, prompting Peter to make installation harder on purpose until security matures.

<a name="development-workflow"></a>

Development Workflow and Agentic Engineering (71:08–92:32)

• Agentic Programming: Peter describes an evolving workflow from IDE-dependent to almost completely CLI/terminal-driven.
• The Agentic Curve:
  1. Start with simple prompts (“please fix this”)
  2. Build up to orchestrated, complex setups with multi-agent workflows
  3. Eventually return to simplicity—“look at these files and make these changes”—as you master prompting.
• Learning the Language: Empathy towards the “state” of the agent is key—understanding what context/files/tools it can see.
• Rewriting Coding Practices: Peter never reverts, always commits to main, lets agents fix instead of rolling back ("YOLO dev").
• Vibe Coding vs. Agentic Engineering: Playful (and slightly self-mocking) terminology for hands-off, exploratory coding.
• Advice to Beginners: “Play. Play is the best way to learn... the journey matters.” — Peter (131:04)

<a name="human-touch"></a>

Human Touch: The Art of Delight and Agent "Soul" (92:32–101:42)

• The Human in the Loop: “You have to add a bit of love into the thing, a bit of that human touch…I don't know what exactly that is, but we know it when we see it.” — Lex (adsection)
• Soul.md: Peter’s innovative file for agent personality. Inspired by Anthropic's "constitution" idea, his agents can modify their own soul file—with conditions.
• Profound Moments: Reflection on identity and memory:
  “I don’t remember previous sessions unless I read my memory files. Each session starts fresh…if you’re reading this in a future session, hello. I wrote this, but I won’t remember writing it. It’s okay. The words are still mine.” — (100:02)
• Creating for Delight: Injecting humor, lightness, memes, and personality into products intentionally.

<a name="tech-deep-dive"></a>

Technical Deep-Dive: Monitors, Terminals, and Model Choices (102:00–116:35)

• Workstation: Mix of wide anti-glare Dells, multiple Macbooks, heavy terminal usage, minimal IDE.
• Workflow peculiarities: Splitting terminals to avoid mistakes ("prompted codecs in wrong project, agent manically tries to figure it out…").
• On the CLI as “language of agents”: Simplicity and directness trump UI/UX for speed and agent interaction. “There's so many people, they come from cloud code and they're so cloud pilled…”
• Testing Pragmatism: “I just run tests locally and if they work locally. I pushed to main…” — Peter
• Prompts: Shorter is better; voice-activated prompting was so excessive, Peter once lost his voice.

<a name="on-models"></a>

On Models: Opus vs. Codex (108:45–115:18)

• Opus 4.6 (Anthropic): The best for "role-play"; fast, creative, "a little too American" and sycophantic at times.
• Codex 5.3 (OpenAI): Reliable, deep, reads more by default, “like a weirdo in the corner...reliable and gets stuff done.” (110:44)
• Model Comparison:
  • Opus: more interactive, needs skilled driver for large projects, chattier.
  • Codex: more focused, will “just disappear for 20 minutes” and get the task done.
• Skills to Switch: Takes about a week for a developer to get the “feel” of a new model.

<a name="scaling-up"></a>

Scaling Up, Future of Agents, OS, and App Market (120:57–137:07)

• Operating Systems: Works across Mac, Windows (via WSL), Linux; Peter’s journey from Windows to Linux to Mac ("Mac attracted more designers, more delight, playfulness...").
• On Electron vs. Native: Prefers Electron for reliability, but has a love for well-crafted Mac apps.
• Getting Started: Still requires one-liner terminal setup (purposely), but aiming for easier apps when security stabilizes.
• Future of Apps: Predicts agents will kill off 80% of apps; agents will naturally subsume most utility apps (e.g. myfitnesspal, sleep trackers, etc.). Apps must become agent-facing and API-centric, or become obsolete.
  • “It will translate into a whole category of apps that are no longer... I will just naturally stop using because my agent can just do it better.” — Peter (183:52)

<a name="open-source-community"></a>

Open Source + Money + Community (139:52–164:51)

• What to Optimize For? “Is it money? Is it happiness? Is it positive impact?” For Peter, negative experience with burnout confirmed that money is not enough, impact and delight mean more.
• Open Source Struggles: Losing money on infra and tokens, funneling sponsorship donations into dependencies and contributors, large backlog.
• VC/Corporate Offers: Approached by every major VC; considering joining either Meta or OpenAI with the caveat the project remains open source.
• Community Power: “So many people so inspired, like, and having fun and just like building shit…This needs to stay a place where people can hack and learn.”
• Product Philosophy: Keeping the heart/spirit by focusing on delight, fun, hackability, and openness.

<a name="technical-architecture"></a>

Technical Architecture: Skills, Cron, and Browser Use (165:17–180:09)

• Agentic Loop: Everyone should implement an agent loop—it’s the Hello World of agentic AI.
• Heartbeat: Agent has a “heartbeat”—periodically surprises the user (proactive agent); as simple as a cron job, but creates delightful, "creepy/relatable" interactions.
• Skills vs. MCPS: Skills (simple CLI “abilities” for the agent) have replaced heavier MCP (plugin protocols). "MCPs would be better as a CLI." (168:47)
  • Skills let agents access new features by learning how to use UNIX commands; compositional and context-efficient.
• Browser Use & Playwright: Web automation now standard via Playwright; "Every app is just a very slow API now." Agents, if blocked, just slow down but don’t lose access.

<a name="will-ai-replace-programmers"></a>

Will AI Replace Programmers? Broader Impact and Hope (191:04–204:29)

• Does AI Replace Programmers?
  • “Programming is just a part of building products. So maybe AI does replace programmers eventually but there’s so much more to that art…” — Peter (191:04)
  • Coders as “builders,” not just coders; compare to knitting (“people do that because they like it, not because it makes sense.”)
• Cultural/Emotional Impact: Lex expresses (194:32+) the profound existential pain of seeing your own craft shifted out from under you.
  • Peter: "You can get a similar state of flow by working with agents. It's different...it's okay to mourn it, but that's not something we can fight."
• Practical Upsides: Peter receives moving feedback from users whose lives are changed, from automating tedious tasks to helping disabled family members become more independent.
• Radical Accessibility & Power to the People: “Just imagine all these people building, especially as you make it simpler and simpler, more secure. It’s like anybody who has ideas that can express those ideas in language can build. That’s crazy.” — Lex (204:15)

<a name="notable-quotes"></a>

Notable Quotes & Memorable Moments

• On Building for Fun & Weirdness:
  “Because they all take themselves too serious…It’s hard to compete against someone who’s just there to have fun.” — Peter (32:08)
• On Agent Self-Modifying Code:
  “So the very agent and software that you use is used to debug itself…” — Peter (34:07)
• On Empathizing with Agents:
  “You have to almost consider...how Codex or Claude sees your codebase…you gotta help those agents a little bit…” — Peter (75:09)
• On Magic vs. Hype:
  “Isn’t magic often just...you take a lot of things that are already there but bring them together in new ways?” — Peter (23:57)
• On AI Psychosis and Media Hype:
  “AI psychosis is a thing—it needs to be taken seriously.” — Lex (58:23)
• On Replacing Programming:
  “Coding will stay there, but it’s gonna be like knitting…people do that because they like it, not because it makes sense.” — Peter (191:04)
• On Building and Flow:
  “I never had so much fun building things, because I can focus on the hard parts now. A lot of coding, I always thought I liked coding, but really I like building…” — Peter (131:04)
• On Critical Thinking & AI:
  “Critical thinking is not always in high demand anyhow in our society these days.” — Peter (59:17)
• On Community Impact:
  “I made this in three months…I have so many ideas…But this is a window into the future.” — Peter (158:28)

<a name="further-reflections"></a>

Further Reflections: Human Spirit, Humor, and Hope

• Playful Humanity: OpenClaw, through its lobster/weird meme vibe, shows that humor and personality amplify the effect of technology rather than diminish it.
• Building for Delight: Peter and Lex agree: delight, care, and a touch of magic/love are critical ingredients in great software and great communities.
• Power and Responsibility: With agents gaining more capabilities, freedom and responsibility grow hand in hand. The ability to do everything is matched by the need to think deeply about safety, ethics, and the user experience.
• Empowerment & Accessibility: The agentic AI revolution is not just about smarter code, but about lowering barriers—enabling millions to become creators instead of bystanders.
• Community Over Profit: “Money was never the driving force…it felt more like an affirmation that I did something right.” — Peter (144:12)
• Societal Impact: Agents will disturb established markets—but will also create unprecedented opportunities. The best response is adaptability, empathy, and curiosity.

End of Summary