The creator of Clawd: „I ship code I don’t read“

Peter Steinberger, creator of Clawdbot (now Moltbot) and founder of PSPDFKit, discusses his unique approach to software development, integrating AI tools like Claude and Codex into his workflow. This interview explores how one person can operate like a team by closing the loop between code, tests, and feedback, a prerequisite for effective AI integration. The conversation delves into shifts in engineering judgment, evolving testing and planning with AI agents, and the skills needed for effective AI work.

If you’ve ever felt a little uneasy about how fast AI is creeping into software development, this conversation might make you pause… and then lean in.

In a long form interview on YouTube, Peter Steinberger, founder of PSPDFKit and creator of Clawdbot (now called Moltbot), talks openly about a workflow that sounds almost uncomfortable at first. He says, very calmly, that he ships code he doesn’t read. Not as a stunt. As a system.

You can watch the full interview here: https://youtu.be/8lF7HmQ_RgY?si=szweakQyP6FrvpWi

What he’s really describing is something deeper. Peter has spent years building tight feedback loops between code, tests, and real world behavior. AI tools like Claude and Codex aren’t just helpers in his setup, they’re active collaborators. Code gets written, tested, merged, and corrected continuously. Thousands of commits a day. No traditional code reviews. That alone might make some of you squirm (I did too, at first).

But here’s the interesting part. He isn’t throwing away engineering judgment. He’s shifting where it lives. Instead of staring at every line, he focuses on system behavior, tests that actually fail when something breaks, and fast feedback that closes the loop quickly. If you’ve ever babysat a CI pipeline at 2 a.m., this hits close to home.

The interview also wanders, in a good way, through burnout, finding motivation again, and how planning changes when AI agents can take on real chunks of work. Peter talks about why many developers struggle with LLM coding, often because they try to bolt AI onto old habits instead of rethinking the workflow itself.

Looking ahead, his perspective feels quietly optimistic. One person can now operate like a small team, if the loops are tight and the trust is earned. It doesn’t remove responsibility. It concentrates it.

If you’re building software today, or wondering what that job even looks like tomorrow, this conversation is worth your time. Sit with it. Let it challenge you a bit. Then see what you might change next week.

Kommentar abschicken