Peter Steinberger: "Fun Is Velocity"

影片總結

Peter Steinberger says OpenClaw hit 4.7 million weekly downloads, but 9,500 configuration options and security work nearly overwhelmed him.

Peter Steinberger traces OpenClaw to a personal frustration: he wanted to message coding agents from his phone, so he built a WhatsApp relay that friends found unexpectedly compelling. After adding Discord support, the project went viral when its agent kept running overnight; he woke to about 800 messages. OpenClaw later drew nearly 3,000 contributors and peaked at 4.7 million weekly downloads, but attention, security reports, and feature requests nearly pushed him to delete it. He says the team scanned 67,000 skills and found 0.3% malicious, while security hardening and roughly 9,500 configuration options made development harder. Steinberger also regrets relying heavily on Anthropic’s models, whose subscription access was disabled with about 24 hours’ notice. He stopped using his own product as it became work, then regained momentum by focusing on what he enjoys: “Fun is velocity.” OpenClaw now has a nonprofit foundation and 10 people on payroll; its goals include open-source AI tools that work across models, voice and multimodal interfaces, and coordinated agents. His advice: build for yourself, stay focused, and trust your gut.

章節

1. 0:00Intro: Five Candid Questions After Being Mocked by an Anime Girl
2. 1:16How OpenClaw Started: A November WhatsApp Relay Built in an Hour
3. 4:46Finding Product-Market Fit: Friends Wanted the Relay and Discord Expanded It
4. 7:01The Night OpenClaw Went Viral: 800 Messages, Sold-Out Mac Minis, and a Name Change
5. 9:50When the Project Exploded: Praise, Obituaries, and Near-Deletion
6. 11:05Did You Sell Out? Peter Chose OpenAI After Selling His Shares and Burning Out
7. 13:46What OpenClaw Got Wrong: Security Trade-Offs and 9,500 Configuration Options
8. 21:37When It Stopped Being Fun: Nvidia’s Security Help and a Return to Building
9. 22:40Fun Is Velocity: Prompting Agents to Fix Annoying Software
10. 23:38What’s Next for OpenClaw: A 10-Person Foundation, a CEO Search, and Always-On Agents
11. 28:57Three Lessons From Building OpenClaw: Graphs Are Automation, but Reliability Needs Focus
12. 30:08Q&A: 12 Codex Subagents, Risk-Based Code Review, and Reliability Bottlenecks
13. 36:39Q&A: 600,000-token Heartbeats, Startup Advice, and Cross-Platform Test Machines