← All speakers

Bio, Work & Ideas

Peter Steinberger

Conference affiliation: OpenAI · 2026

Peter Steinberger is the creator of OpenClaw, an open-source personal-agent platform, and an OpenAI engineer working to bring autonomous assistants to a broader audience. Previously, he co-founded PSPDFKit, building document infrastructure for developers before turning to software agents that operate across people’s devices, applications, and communication channels.

Steinberger established himself in the Apple development ecosystem through PSPDFKit’s PDF viewing, annotation, signing, and collaboration tools. His work bringing SwiftUI into a production PDF framework reflects a longstanding preoccupation with integrating emerging technology into reliable developer products. In 2021, PSPDFKit received more than €100 million in strategic investment, and Steinberger and fellow co-founder Martin Schürrer stepped back from their full-time roles.

His subsequent open-source projects included CodexBar, a coding-agent usage monitor, and Peekaboo, which enables agents to capture and interact with macOS interfaces. OpenClaw expanded that ambition into a personal assistant capable of using tools, connecting to messaging platforms, and working across different models and computing environments.

In February 2026, Steinberger announced that he was joining OpenAI while keeping OpenClaw open and independent. The OpenClaw Foundation gives the project a governance structure separate from his employer, with maintainers and contributors drawn from multiple companies.

  • Personal agents should remain under user control. Steinberger prioritizes personal-data ownership and model choice: assistants should work close to their users’ devices and existing communications without requiring a centralized intermediary to control their entire digital lives. He envisions agents moving between computers, messaging applications, and connected environments while preserving boundaries between personal and workplace information.
  • Agent security depends on actual exposure. He treats prompt-injection risk as a structural consequence of combining private data, untrusted inputs, and outbound communication. His approach emphasizes permission boundaries, sandboxing, supply-chain security, and deployment-specific risk; OpenClaw’s security guidance distinguishes realistic threats from scenarios requiring unsafe configurations.
  • Engineering judgment becomes more valuable as code gets cheaper. Steinberger’s approach to agent-assisted development favors direct experimentation, rapid feedback, architectural discipline, and attention to a change’s wider effects. He resists fully predetermined software factories: product direction, interface details, and deciding which features to reject still require human taste.
  • Persistent agent orchestration should conserve human attention. His workflow has progressed from supervising multiple coding sessions to directing a durable manager agent that delegates tasks, responds to events, coordinates implementation and review, and escalates meaningful decisions.

Talks by Peter Steinberger

2 talks

References