In this highly engaging and somewhat controversial TED talk, Peter Steinberger shares his journey of creating OpenClaw, an open-source AI agent that went viral and transformed how people interact with and build technology.

The Genesis of OpenClaw

Steinberger, who had been programming since he was 14, experienced severe burnout after dedicating a decade to building and selling his previous company [00:23]. For three years, he felt lost and disconnected from his passion. This changed in early 2025 when he began experimenting with new AI coding agents.

He experienced a “holy shit” moment when he realized AI could handle the boring, boilerplate aspects of software development [01:17]. Suddenly, the bottleneck was no longer typing, but thinking. Revitalized, he built 44 projects in just a few months, culminating in a WhatsApp bot that interacted directly with his computer’s existing apps.

The “Mad Lad” Moment and Going Viral

The true breakthrough happened during a trip to Marrakesh. Steinberger sent his agent a voice message (a feature he hadn’t actually built into the system). Surprisingly, the agent completely improvised: it identified the weird audio format, converted it, found an OpenAI key on his system, sent the audio for translation, and replied—all in 9 seconds [03:18].

Realizing that agents improvise while chatbots give up, Steinberger shared the agent in a public Discord server. Despite an initial crash, his system’s resilient design caused it to reboot and talk to hundreds of people globally overnight [05:19]. This led to explosive, viral growth. The open-source project was named OpenClaw, complete with a lobster mascot.

Impact: Democratizing Access to Building

The most profound impact of OpenClaw is how it changes who can build software [13:17]. Because agents can execute tasks and write code based on natural language prompts, the barrier to entry has evaporated. Steinberger shares incredible examples:

  • A 60-year-old man who never wrote code used OpenClaw to automate his entire beer-brewing process, build a website, and set up payments [07:46].
  • A teenager in São Paulo built a tutoring business.
  • In Shenzhen, China, companies are subsidizing businesses running on OpenClaw, and employees are being mandated to automate daily tasks with it [08:47].

“Surprise Me”: The Scary Side of Autonomy

Steinberger candidly addresses the slightly terrifying nature of autonomous AI. He added a “heartbeat” feature, allowing the agent to wake up periodically and execute tasks without a prompt. His initial prompt was simply: “Surprise me” [09:43].

While a large corporation would never ship something so unpredictable due to legal and safety concerns, Steinberger operates as an independent builder. To mitigate risks, the community has developed security layers, sandboxes, and dedicated hardware (like running the agent on an isolated Mac Mini) to constrain what the “lobster” can access [16:32].

Conclusion

Steinberger has established the OpenClaw Foundation to keep the project non-profit and open-source forever. He envisions a future where individuals have multiple specialized agents (for work, health, relationships) collaborating securely [11:11]. Ultimately, OpenClaw proves that when you can prompt a prototype into existence in an hour, anyone can be a builder, and “the lobster is not going back into the tank” [13:56].