Steinberger began with a simple link between WhatsApp and Claude Code so he could check on his computer from his phone. The project grew into an assistant that can work through messaging apps, and he describes it as a new kind of operating system for a powerful AI that follows people through life. In Morocco, it read a bug report sent as an image, fixed the code, committed the change, and replied to the person who reported it.
How OpenClaw's Creator Uses AI to Run His Life in 40 Minutes
Speaker
Peter Steinberger, creator of OpenClaw, joins Peter Yang on the Peter Yang channel to discuss building a messaging-based AI assistant and using AI to code and manage everyday tasks.
When Steinberger sent an unsupported voice message, the assistant identified the audio file, converted it, found an API key, and got a transcript. He says assistants can also access email, calendars, files, cameras, home devices, and browsers, with examples including flight check-in and controlling lights and speakers. That reach can create mistakes or dangerous access, so he cautions that users need boundaries around what the assistant can do.
Steinberger describes moving from years of Apple development to building a web app in TypeScript. AI helped him get past unfamiliar syntax while he applied his experience with system design and technical judgment. He also walks through connecting the assistant to models and messaging apps, noting that it can be installed on macOS, Linux, and Windows.
Steinberger warns that elaborate agent managers and long-running loops can become distractions that produce code without a clear product vision. He prefers discussing a feature with AI, trying it, and refining the idea based on what he sees, rather than specifying everything upfront. He says engineers should practice communicating clearly with models, ask questions when useful, and keep human taste involved.
Steinberger uses conversations and support questions to find product pain points, including a daily review of scraped help discussions. He says his coding setup is simple: several terminals and repository checkouts, with him testing completed changes before pushing them. He encourages people to explore use cases themselves and learn by making mistakes.
Takeaways
- Start with a recurring task in your life and see whether an assistant can handle it through tools you already use.
- Give an AI assistant access only to the accounts, files, and devices needed for its tasks.
- Use your engineering experience to guide AI across unfamiliar languages and frameworks.
- Iterate with AI as you inspect the product; a detailed plan alone cannot supply product judgment.
- Build tools around real user pain points, and learn by trying and refining them.
Mentioned in this talk
Search inside this talk
What was claimed 8
- builtPeter Steinberger built
OpenClaw
“the creator of OpenClaw”
0:00inState of the ClawYouTube Verified7 sources· Talk, Web· archived
- feature
OpenClaw: Connects through messaging apps and can use tools on a computer to perform tasks
“an AI assistant that you can chat with in your messaging apps to get stuff done”
YouTube Unverified1 source· Talk
- integrates with
“I've got Claude set up on my Telegram messaging app”
0:47inFull Tutorial: Set Up Your 24/7 AI Employee in 20 MinutesYouTube Verified2 sources· Talk
- metric
OpenClaw: About 300,000 lines of code
“I think it's 300,000 lines of code”
YouTube Unverified1 source· Talk
- integrates with
“The small thing was hooking up WhatsApp to Claude Code.”
YouTube Verified2 sources· Talk
- feature
OpenClaw: Can be installed on macOS, Linux, and Windows
“It works on macOS, Linux, and Windows”
YouTube Unverified1 source· Talk
- usesPeter Steinberger uses
Codex
“I love Codex because it looks through more files before deciding what to change.”
18:53inOpenClaw Creator: Why 80% Of Apps Will DisappearYouTube Verified3 sources· Talk
- feature
OpenClaw: Has persistent memory and can learn about the user
“it has persistent memory. It will learn about you and update itself”
YouTube Unverified1 source· Talk











