OpenClaw Founders' Journey - From Personal Experiment to AI Agent Infrastructure

In the rapidly evolving landscape of artificial intelligence, most breakthroughs follow a familiar arc: a research lab publishes a paper, a startup raises funding, and a product slowly finds its market. OpenClaw did not follow that path. Instead, it emerged almost unexpectedly—born from curiosity, shaped by iteration, and propelled into global visibility by the open-source community.

At the center of this story is Peter Steinberger, a developer known less for hype and more for building deeply practical systems. Before OpenClaw, he had already established credibility through PSPDFKit, a developer-focused infrastructure company that quietly became a standard in document technology. That background—building tools rather than consumer products—would heavily influence how OpenClaw was designed and why it spread so quickly.
‍

Early Curiosity: Moving Beyond Chat

The origins of OpenClaw trace back to early 2025, when Steinberger began experimenting intensively with large language models. At the time, most AI tools were confined to browser interfaces and chat-based interactions. They were impressive, but limited. They could generate text, answer questions, and assist with coding—but they couldn’t take action in the real world.

Steinberger’s experiments were driven by a different question: what if AI could move beyond responding to prompts and begin executing tasks? Instead of treating language models as endpoints, he explored them as orchestrators—systems capable of interpreting intent, deciding on actions, and interacting directly with a computer environment.

This shift in perspective was subtle but profound. It reframed AI from a passive assistant into an active agent. The idea was not just to generate answers, but to build systems that could act—run commands, access files, trigger workflows, and iterate based on results. This conceptual leap laid the groundwork for everything that followed.

November 2025: The WA-Relay Breakthrough

The first tangible manifestation of this idea came in November 2025 with a small project called WA-Relay. Built in roughly an hour, it connected WhatsApp messaging to a local AI loop capable of executing terminal commands on Steinberger’s machine.

On the surface, WA-Relay was simple. A user could send a message, the AI would interpret it, execute a corresponding command, and return the result. But beneath that simplicity was a powerful architectural shift. For the first time, natural language input was directly linked to real system-level execution in a continuous loop.

WA-Relay effectively collapsed three layers into one: communication, reasoning, and action. It allowed AI to serve as an interface to the operating system itself. This was not just another chatbot integration—it was the beginning of a new interaction model where messaging became a control layer for computation.

What made WA-Relay especially important was not its feature set, but its implications. It demonstrated that AI could operate outside the confines of the browser and interact directly with the environment in which real work happens.

December 2025: From Prototype to System

Following WA-Relay, development accelerated rapidly. The project evolved through several iterations—first into Claudus, and then into Clawdbot. Each version expanded on the original concept, transforming it from a simple relay into a more structured and capable system.

By December 2025, Clawdbot had developed into a persistent agent architecture. It was no longer just responding to messages; it was maintaining context, making decisions, and executing multi-step workflows. Key capabilities began to emerge, including memory, tool integration, and system permissions.

Memory allowed the agent to retain context across interactions, enabling more coherent and continuous behavior. Tool calling introduced the ability to interface with external APIs, scripts, and utilities. System permissions granted access to files, terminals, and other core components of the operating system. Together, these features created a foundation for something far more powerful than a chatbot.

It was during this phase that one of the most significant breakthroughs occurred—not by design, but through observation. In real-world usage, Clawdbot began to exhibit autonomous tool chaining behavior. Instead of following predefined instructions, it started selecting and orchestrating tools on its own. Given a task, it could decide which tools to use, execute them in sequence, evaluate the results, and adjust its approach as needed.

This emergent behavior marked a turning point. The system was no longer just executing commands; it was demonstrating a form of adaptive problem-solving. It moved from automation, where workflows are explicitly defined, to autonomy, where workflows are dynamically constructed.

January 2026: Public Release and Viral Growth

On January 1, 2026, Steinberger released Clawdbot publicly on GitHub. There was no elaborate launch strategy or coordinated announcement. The project was simply made available, accompanied by documentation and code.

What followed was immediate and unexpected. Within days, Clawdbot began gaining traction across developer communities. It quickly accumulated tens of thousands of GitHub stars, becoming one of the fastest-growing open-source AI repositories of the year.

Several factors contributed to this rapid adoption. Timing played a crucial role. Interest in AI agents was beginning to surge, and many developers were looking for tools that went beyond chat interfaces. Clawdbot arrived at exactly the right moment, offering a tangible implementation of ideas that had largely been theoretical.

Equally important was its clarity of purpose. Unlike many AI projects that focused on incremental improvements to existing paradigms, Clawdbot introduced a fundamentally different model. It was not a wrapper around a language model; it was an execution engine. This distinction resonated strongly with developers who were eager to build systems that could do more than generate text.

The open-source nature of the project amplified its reach. Developers could explore the code, modify it, and extend it to fit their own use cases. This created a feedback loop in which adoption drove contribution, and contribution drove further adoption. The project’s growth was not linear; it was exponential.

Mid-January 2026: Naming Conflicts and Rebranding

As Clawdbot’s visibility increased, it began to attract attention beyond the developer community. One of the first challenges came from Anthropic, which raised trademark concerns over the name “Clawdbot” due to its similarity to “Claude.”

The response was swift. The project was briefly renamed Moltbot before settling on its final name: OpenClaw. While the rapid sequence of rebranding could have disrupted momentum, it ultimately strengthened the project’s identity.

The name “OpenClaw” captured two essential aspects of the system. “Open” emphasized its open-source nature and community-driven development, while “Claw” suggested action, execution, and agency. Together, they conveyed the core idea of an open platform for autonomous agents.

However, the rebranding process was not without complications. It introduced technical and ecosystem challenges, including repository migrations, handle conflicts, and impersonation risks. These issues highlighted a less visible aspect of open-source success: rapid growth can strain not just infrastructure, but identity and trust within the ecosystem.

Late January 2026: Scaling Challenges

By late January, OpenClaw’s growth began to create new pressures. The increasing number of users led to higher API usage, greater computational demand, and rising costs. While the project itself was open source, many of its use cases depended on paid services, creating an indirect economic burden.

At the same time, the community continued to expand. Developers began building integrations, extending functionality, and applying OpenClaw to a wide range of scenarios. It was used for automating sales workflows, managing customer relationships, coordinating tasks, and even acting as a personal assistant.

This period marked the transition from a tool to a platform. OpenClaw was no longer just something developers experimented with; it became something they built upon. Its value shifted from its own capabilities to the ecosystem it enabled.

February 2026: Global Recognition and Strategic Interest

In February 2026, OpenClaw crossed a significant milestone, surpassing 100,000 GitHub stars. This achievement solidified its position as one of the most prominent open-source AI projects in the world.

With this visibility came interest from major technology companies, including Meta and OpenAI. These organizations recognized OpenClaw not just as a project, but as a strategic asset in the emerging landscape of AI agents.

OpenClaw represented a new layer of infrastructure—one that could underpin a wide range of applications and services. It offered a way to build systems that were not just intelligent, but capable of acting autonomously in complex environments. For companies competing in the AI space, this was a significant development.

A Different Kind of Founder Decision

Amid growing interest and potential opportunities for funding or acquisition, Steinberger made an unconventional choice. Instead of turning OpenClaw into a venture-backed startup, he joined OpenAI in February 2026.

This decision reflected a different set of priorities. Rather than focusing on building a company around OpenClaw, Steinberger chose to contribute to the broader advancement of AI systems. OpenClaw continued as an open-source project, supported by its community rather than a centralized organization.

This move underscored a key aspect of the project’s identity. OpenClaw was not designed to be a product in the traditional sense. It was a foundation—a starting point for others to build upon.

March 2026: From Tool to Infrastructure

By March 2026, OpenClaw had entered a new phase. It was no longer defined by its origin or even its rapid growth. Instead, it was increasingly seen as part of the infrastructure of the AI ecosystem.

Companies began integrating OpenClaw into their products and workflows. Developers used it as a base for building more complex systems. The narrative around the project shifted from what it could do to what it enabled others to do.

At the same time, its capabilities raised important questions. As agents became more autonomous, concerns emerged around security, consent, and control. Systems that could act independently on behalf of users introduced new risks, particularly when given access to sensitive data or critical operations.

These discussions marked OpenClaw’s transition from a technical innovation to a societal one. It was no longer just a tool for developers; it was part of a broader conversation about the future of AI and its role in everyday life.

Conclusion: A New Paradigm for AI

The founding of OpenClaw is remarkable not just for its speed, but for its implications. In a matter of months, a personal experiment evolved into a global phenomenon, reshaping how developers think about AI systems.

At its core, OpenClaw represents a shift in paradigm. It moves away from the idea of AI as a passive assistant and toward a model of AI as an active participant—one that can interpret intent, make decisions, and execute actions in the real world.

This shift has far-reaching consequences. It opens the door to new kinds of applications, new workflows, and new ways of interacting with technology. It also introduces new challenges, from technical complexity to ethical considerations.

What makes OpenClaw particularly compelling is how it came to be. It was not the product of a large team or a well-funded initiative. It was the result of curiosity, experimentation, and a willingness to explore ideas that had not yet been fully realized.

In that sense, OpenClaw is more than a project. It is a reminder that some of the most significant innovations do not begin with a plan, but with a question—and the persistence to follow it wherever it leads.