OpenClaw Next Week Update: Support for Claude Code and OpenAI Codex, Founder Steinberger Still Balancing Code Writing

Peter Steinberger’s OpenClaw announced support for Claude Code and OpenAI Codex CLI plugins next week, while also splitting the core engine to reduce package size—but the fact that this former PSPDFKit founder is still managing two companies at once remains the real risk factor to watch.
(Background: Cursor launches background AI agent: automatically fixes bugs, builds features, developers just give commands)
(Additional context: Anthropic releases Claude Code: an end-terminal AI assistant that writes code directly in the command line)

Table of Contents

Toggle

• Feature list: everything you’d expect, but nothing surprising
• The real issue: can one person run two companies at the same time?
• Calm market positioning analysis

Another AI coding tool announced an update via tweet on Sunday. This time it’s OpenClaw—an open-source AI coding agent created by former PSPDFKit founder Peter Steinberger. He posted a series of tweets on X detailing the upcoming features list for next week.

Honestly, the list itself isn’t surprising. Support for Claude Code plugins, integration with OpenAI Codex CLI, compatibility with ACP (Agent Communication Protocol)—these are all must-have features for any product aiming to survive in the AI coding tools market, not unique differentiators.

Feature list: everything you’d expect, but nothing surprising

According to Steinberger’s tweets, the core updates next week include:

• Claude Code plugin support—enabling OpenClaw to function as an extension within the Anthropic ecosystem

• OpenAI Codex CLI plugin—compatibility with OpenAI’s command-line development tools

• ACP protocol support—enabling cross-agent communication, the foundational infrastructure for multi-agent architectures

• Agent interruption mechanism fix—solving issues with gracefully stopping agents mid-execution

• Splitting core engine into standalone modules—reducing installation size and improving modularity

Together, these features roughly define OpenClaw’s positioning: not to replace full IDEs like Cursor or Windsurf, but to serve as a lightweight, embeddable, open-source agent engine. Strategically, this isn’t a bad approach, but execution is the key concern.

The real issue: can one person run two companies at the same time?

Steinberger is still a co-founder of PSPDFKit (now renamed Nutrient), a company with over 100 employees serving thousands of enterprise clients. Meanwhile, he’s pushing forward with OpenClaw, a fast-iterating open-source project.

The pace of open-source AI tools is measured in weeks. Claude Code updates every few days; Cursor’s iteration speed is similarly rapid. In such an environment, it’s reasonable to worry whether a part-time founder can maintain enough involvement.

Of course, Steinberger’s technical reputation in developer tools is unquestioned—PSPDFKit is one of the few independent companies to survive over a decade in the B2B SDK market. But technical skill and bandwidth are two different things.

Calm market positioning analysis

Currently, the AI coding tools market roughly divides into three layers:

• First layer: full IDEs—Cursor, Windsurf, GitHub Copilot Workspace

• Second layer: CLI agents—Claude Code, OpenAI Codex CLI, Aider

• Third layer: embeddable engines—the position OpenClaw is trying to occupy

The idea behind the third layer is “let others use my engine in their tools,” which relies on network effects within an ecosystem. OpenClaw has some stars on GitHub, but it’s still far from establishing a developer ecosystem.

It’s worth noting that Anthropic’s Claude Code has already opened up plugin systems, and OpenAI’s Codex CLI is rapidly expanding its plugin ecosystem. When platform providers start building their own ecosystems, third-party engines will face further pressure.

This isn’t to say OpenClaw has no value—open-source alternatives always have a market, especially for enterprise users with sensitive data. But “open source” itself isn’t a moat; sustained high-quality maintenance is, and that circles back to bandwidth constraints.