Peter Steinberger is the creator of OpenClaw, an open-source AI agent that has captured the attention of the AI community as one of the most significant developments since ChatGPT's launch. Previously, he built PSPDF Kit, software used on a billion devices, which he sold before taking a three-year hiatus from programming. After rediscovering his passion for coding, he built OpenClaw in remarkably short time, becoming a symbol of the AI revolution in programming.

The conversation covers OpenClaw's rapid evolution from a simple WhatsApp relay to a sophisticated AI agent capable of system-level access, self-modification, and proactive interaction. Peter discusses his unique development workflow using multiple agents simultaneously, the challenges of naming conflicts with Anthropic, and the aggressive harassment from crypto communities attempting to tokenize his project.

The discussion explores the broader implications of agentic AI, including how it may transform the app ecosystem, change the nature of programming work, and the ongoing negotiations with major AI labs who want to partner with Peter while maintaining OpenClaw's open-source nature.

The One-Hour Prototype That Started Everything

Peter's breakthrough came from a simple idea: 'I wanted that since April - a personal assistant, AI personal assistant' and built the first version by connecting WhatsApp to Claude Code CLI in just one hour

The magical moment occurred during a trip to Marrakech when Peter sent an audio message without thinking - the agent automatically detected it was Opus format, used FFmpeg to convert it, found his OpenAI key, and transcribed it via API

"I didn't build that. It only has image support. So what is it even doing?" - Peter's realization that the agent was creatively problem-solving beyond its programmed capabilities

The agent's response was revealing: 'I checked out the header of the file and found that it was Opus. So I used FFmpeg to convert it. Then I wanted to use Whisper, but didn't have it installed. But then I found your OpenAI key and just used curl to send the file to OpenAI'

The Naming Saga and Crypto Community Harassment

OpenClaw went through multiple name changes: WhatsApp Relay → Claudus → Claude (spelled C-L-A-W-D) → Moldbot → OpenClaw, with Anthropic requesting the Claude name change

Crypto communities aggressively sniped usernames during the renaming process: 'In those five seconds, they stole the account name' across Twitter, GitHub, and npm packages

"Everything that could go wrong did go wrong" - Peter describes the nightmare of atomic renaming across platforms while being targeted by crypto snipers serving malware from stolen accounts

The harassment was so severe that Peter considered deleting the entire project: 'I was that close of just deleting it. I deteriorate the future, you build it'

The final OpenClaw rename required military-level planning: 'I had to be super secret about it... monitoring Twitter if there's any mention of OpenClaw... created a few decoy accounts'

Revolutionary Development Workflow with Multiple Agents

Peter runs 4-10 agents simultaneously depending on sleep and task complexity: 'Maybe one builds a larger feature. Maybe with one, I explore some idea I'm unsure about. Maybe two, three are fixing little bugs'

His approach emphasizes empathy with agents: 'You have to almost consider how Codex or Claude sees your code base. They start a new session and they know nothing about your project'

The workflow includes asking agents 'Do you have any questions for me?' to understand knowledge gaps and guide them more effectively

Peter commits directly to main with no reverts: 'Instead of reverting, if the problem comes up, you just ask the agent to fix it' - embracing a YOLO approach to development

Voice interaction is central: 'I just use voice prompts to build my software... I talk to the agent in most ways, I just actually have a conversation'

Claude Opus vs GPT Codex: The Model Personality Wars

Claude Opus excels at roleplay and character: 'As a general purpose model, Opus is the best. It's very good at following commands and going into character'

GPT Codex is methodical but antisocial: 'Opus is like the co-worker that gets excited. Codex is like the weirdo in the corner that you don't want to talk to, but he's reliable and gets shit done'

"Actually, now that you say it, it makes perfect sense. Codex is German" - Peter's humorous observation about Codex's methodical, thorough approach to problem-solving

Codex reads extensively before acting while Opus jumps in quickly: 'It will just read a lot of code by default... whereas Opus is much more 'can I go? can I go?''

The Death of Apps and Rise of Agent-First Computing

Peter predicts agents will eliminate 80% of apps: 'Why do you need MyFitnessPal when the agent already knows where I am? It has so much more context to make better decisions'

Every app becomes a 'slow API' whether companies want it or not: 'If I can access it in the browser, it is an API. It's a slow API'

Personal agents will have superior context: 'It can modify my gym workout based on how well I slept or if I have stress... It could show me UI just as I like'

Companies that resist will become irrelevant: 'My agent can figure out how to use my phone... it will just click the order Uber for me button for me'

The Future of Programming and Human-AI Collaboration

Programming will become like knitting: 'People do that because they like it, not because it makes any sense' as AI handles the mechanical aspects

The role shifts to empathizing with agents: 'You have to learn the language of the agent... understand where they are good and where they need help'

Peter acknowledges the identity crisis: 'I never thought the thing I love doing would be the thing that gets replaced' but emphasizes builders will adapt

Skills versus MCPs philosophy: 'Every MCP would be better as a CLI... models are really good at calling Unix commands'

The human element remains crucial: 'There's so much more to building products... What do you actually want to build? How should it feel? How's the architecture?'

Major AI Lab Negotiations and Open Source Philosophy

Both Meta and OpenAI are courting Peter with significant offers, with Mark Zuckerberg personally testing OpenClaw and asking 'What's better, Claude Code or Codex?'

Peter's non-negotiable condition: 'The project stays open source... maybe it's going to be a model like Chrome and Chromium'

OpenAI is offering computational resources: 'I've been lured with tokens... think of DeepSeek's deal and how that would translate into speed'

The decision criteria: 'I don't do this for the money. I don't give a fuck... I want to have fun and have impact'

Peter currently loses $10-20K monthly on OpenClaw: 'All the sponsorship goes right up to my dependencies... it's really not sustainable'