Peter Steinberger:
From PSPDFKit about Burnout
to OpenAI

It was a Tuesday evening in November 2025. Peter Steinberger was sitting at his computer in Vienna and carried out an experiment that he himself described as "completely unspectacular" describes. He connected WhatsApp to the Claude API. Sent a test message. The AI ​​responded. The first version was ready in an hour.

He couldn't have predicted what happened next. In less than three months This weekend project would collect 247,000 GitHub stars, the AI community of Employ Silicon Valley to Beijing and get Sam Altman to hire him to publicly describe him as a “genius”.

This is the story of the man whose name has long been bigger than any of the projects that made him famous.

Peter Steinberger was born on May 22, 1986 in Upper Austria. He grew up in the region on, visited the HTL Braunau – a technical college – and then moved to Vienna, where he attended the Vienna University of Technology studied.

At the TU Vienna, Steinberger discovered his passion for Apple platforms. He was one of the earliest iOS developers in Austria and worked as a freelancer and taught iOS and Mac development at university. To this day, his online alias is: steeped – a name, which has almost more weight in the developer community than its commoner.

In 2011, he received a job offer at the WWDC party in San Francisco. The temptation was great: Silicon Valley, startup culture, elite developers. He accepted. And waited over six months for his work visa.

Those six months changed everything. Instead of waiting idly, he built something.

What Steinberger built in this waiting period was PSPDFKit – a software development kit for displaying and processing PDF documents on mobile devices. The idea was simple: PDF handling was solved on iOS Problem, but a poorly solved one. Steinberger wanted to do better.

The growth was organic and uncompromising. Steinberger optimized obsessively on Developer Experience – fast integration, excellent documentation, immediate support. Companies that once integrated PSPDFKit stayed. The market grew with smartphone adoption.

Thirteen years. No external investor, no venture capital, no growth rounds. PSPDFKit was fully bootstrapped – one of the rare examples where a software product in the B2B sector without Borrowed capital grew to global market leadership.

What looked like a glamorous exit on the outside was on the inside another story. Steinberger has spoken openly in interviews about what thirteen years of bootstrapping really means: working almost every weekend, up to 100 hours per week, a permanent burden on your shoulders of the founder, who is ultimately responsible for everything.

After withdrawing from PSPDFKit, what followed was what many successful founders know: but only a few publicly name: Burnout. Difficult. Persistent. Existential.

Steinberger largely disappeared from public view. For three years. No products. No talks. No GitHub activity. He moved several times - Vienna, London, San Francisco - and tried to find his place in a world in which he suddenly no longer belongs was a CEO with a thousand responsibilities.

The emptiness that remains after a decades-long hustle, is well documented in founder biographies. At Steinberger It was particularly deep because PSPDFKit wasn't just his company - it was his core identity.

In 2024, Steinberger began trying out AI-powered development. What he found fundamentally changed his relationship with programming. For the first time in years, coding was fun again.

The term that became established for his approach: Vibe coding. Not debugging stack traces for hours. Not manual writing from Boilerplate. Instead: communicate the outcome to the agent, keep the architecture in mind and let the AI do the rest.

In the Pragmatic Engineer Podcast he talked about his philosophy in one sentence, which polarized the developer community:

The sentence sounds provocative. But the explanation behind it is nuanced: Steinberger had learned to deal with code in 13 years of PSPDFKit, which did not meet his own expectations - because 70 employees just write differently. This ability to let go of perfection and focusing on function made it the ideal one “agentic engineer": someone who doesn't fight with AI, but dances.

He began sharing his findings publicly – on steipete.me, in podcasts, in GitHub repositories for AI workflows, prompt engineering and developer productivity.

In November 2025, Steinberger published a small project on GitHub. Not a big announcement. No launch video. Simply code. Clawdbot – a connection between WhatsApp and Anthropics Claude API – was online.

The project initially lay dormant unnoticed. Steinberger continued to build, added features, integrated more messaging platforms. The agent could read emails, manage calendars, organize files, Execute commands on the local machine - all privately, on your own hardware, without cloud dependency.

What differentiated OpenClaw from ChatGPT, n8n or other AI tools: It was not SaaS, not subscription, not cloud. It was an agent who really acted. Not only answered.

Steinberger himself describes the principle behind OpenClaw as “Closing the Agentic Loop”: The agent doesn’t just think – he acts in the real world, observes the effects of his actions, learns from it and changes his own behavior. Without human Intervention in every step.

The renaming – from Clawdbot to Moltbot to OpenClaw – and the chaos around the handle sniping crypto scam generated worldwide coverage on CNBC, Reuters, Engadget and Computerworld. (The complete one Read naming story →)

The ecosystem grew exponentially: Custom Skills, an agent marketplace (ClawHub), Enterprise Forks, an AI social network parody (Moltbook, 1.5 million agent accounts in five days) and finally NVIDIA, which is at GTC 2026 announced its own OpenClaw fork: NemoClaw.

In early February 2026, Steinberger began visiting the AI ​​labs in San Francisco. Not as a visitor – as a candidate. There was interest in the hype surrounding OpenClaw awakened at the highest level. The offers were on the table.

It wasn't a classic acqui-hire. OpenClaw was not purchased. The project remained open source, under MIT license, in an independent foundation. Steinberger signed as Engineer – with a clear mandate: the next generation of personal AI agents for all people.

Steinberger's own words were shorter but programmatic: “I'm joining OpenAI to work on bringing agents to everyone. OpenClaw will move to a foundation and stay open and independent."

Why OpenAI and not Meta, the similar or perhaps higher financial one should have made offers? Steinberger gave an answer in his blog, which reveals a lot about his character:

The analysis community agreed: OpenAI didn't just win with Steinberger a developer. It gained proof that personal AI agents are not only possible, but already viral and ready for production. His wealth of experience - from the developer experience at PSPDFKit to Agentic engineering philosophy at OpenClaw – is for OpenAIs Agent roadmap a direct building block.

Little has changed formally for OpenClaw itself: the project lives in one independent foundation, OpenAI supports it as an official partner, and the MIT license remains untouched. The community – now Thousands of active contributors - continuing to develop.

What this story means for companies

Peter Steinberger's biography is more than an inspiring founder's story. It shows where AI-supported software development is heading: away from large teams, towards small, agentically operated units – provided the governance is right.

OpenClaw is proof that autonomous AI agents ready for production are – at least in technically skilled hands. For medium-sized businesses this means: If you don't understand what agents can do today, you will in two to three years be structurally disadvantaged.

At the same time, the security incidents surrounding OpenClaw show (1,000+ unsecured instances with exposed API keys, 341 malicious Skills on ClawHub with a 12% contamination rate) that enterprise deployments need professional support. This is exactly where Clawgency comes in.

FAQ: Questions about Peter Steinberger