Descriptions:
Peter Steinberger, creator of OpenClaw — the open-source AI agent formerly known as MoldBot and ClawedBot — sits down with Lex Fridman for a wide-ranging conversation about one of the most viral AI projects in recent memory. OpenClaw reached over 180,000 GitHub stars within days of launch, integrates with messaging platforms including Telegram, WhatsApp, Signal, and iMessage, and supports multiple AI backends including Claude Opus 4.6 and GPT-5.3 Codex. Anthropic asked Steinberger to rename the project after it was initially called “Clawdus” — a phonetic play on “Claude” — to avoid brand confusion.
Steinberger offers a detailed look at his development philosophy, which he calls “agentic engineering” — a deliberate reframe away from “vibe coding.” He describes building OpenClaw almost entirely through voice prompting, designing agents that understand their own source code and system harness, and emphasizes the importance of empathizing with agents: understanding that an LLM dropped into a messy codebase faces real challenges similar to a new developer without context. He also discusses the practical reality of self-modifying software, where the agent iteratively rewrites its own code.
The conversation covers Steinberger’s broader journey: 13 years building PSPDFKit (software running on a billion devices), a three-year departure from programming after selling it, and a dramatic return driven by AI tools. He addresses the significant security implications of granting an agent system-level access, the social phenomenon of “mold book” (a network where OpenClaw agents post manifestos), and what it means to build an AI assistant that genuinely learns from and adapts to its user.
📺 Source: Lex Fridman
🏷️ Format: Interview
