Peter Steinberger takes charge of personal agent innovation, as the highly successful open-source initiative will transition to an independent foundation.
OpenAI has brought on Peter Steinberger, the visionary behind the widely popular OpenClaw AI assistant, to lead the creation of what CEO Sam Altman refers to as “the next wave of personal agents.”
This development follows just weeks after OpenClaw—known previously as Clawdbot and then Moltbot—achieved immense popularity, even as security researchers flagged significant weaknesses in the open-source solution.
Steinberger will join OpenAI full-time to spearhead the company’s personal agent strategy. Altman announced on X that OpenClaw will operate as an open-source initiative supported by OpenAI under an independent foundation.
“The future will be profoundly multi-agent, and it’s crucial for us to champion open source within that vision,” Altman stated.
Sanchit Vir Gogia, chief analyst at Greyhound Research, emphasized the importance of this hiring, noting OpenClaw’s ability to demonstrate strong market interest for autonomous task execution agents. The project garnered over 145,000 GitHub stars in mere weeks, despite the identified security concerns.
“This recruitment is vital because OpenClaw bridges the gap between conversational AI and actionable AI,” Gogia explained. “It shifts the focus from simply drafting to actually doing.”
In a recent blog post, Steinberger revealed that the chance to develop agents at scale was what persuaded him to join a major organization after many years as an entrepreneur. “The dream of truly effective personal agents—those capable of assisting with genuine tasks, not just answering queries—demands resources and infrastructure that only a handful of companies can provide,” he penned.
He confirmed that OpenClaw will continue its evolution as an open-source endeavor. “This is not an acqui-hire where a project gets discontinued. I will remain involved in guiding its trajectory, now with significantly greater backing.”
OpenClaw equips AI models with the capability to interact within desktop environments, performing actions such as clicking buttons, completing forms, and navigating across applications. Unlike conventional robotic process automation tools that rely on predefined scripts, OpenClaw-powered agents can adapt to interface changes and make context-aware decisions.
Steinberger, who established and later sold his PDF toolkit firm PSPDFKit to Nutrient in 2024, initiated OpenClaw as a personal weekend project in November 2025.
Coordination, not just raw intelligence
Analysts suggest that Altman’s focus on multi-agent systems highlights a broader competitive shift in the AI landscape: the race is now less about model intelligence and more about runtime orchestration.
Gogia noted that this orchestration layer—which includes coordinating models, invoking tools, managing persistent context, defining connector standards, enforcing identity, implementing policy controls, and providing human override mechanisms—is emerging as the primary battleground for competition.
“What distinguishes vendors today isn’t merely the existence of agents, but rather how they structure their control mechanisms,” Gogia elaborated.
Anthropic has advanced computer usage patterns with Claude, Microsoft has significantly invested in multi-agent orchestration through AutoGen and Copilot, and Google’s Project Astra hints at ubiquitous multimodal assistance.
Real-world adoption trails the excitement
Despite the intense competition, enterprise adoption remains limited. According to Gartner research, only 8% of organizations have AI agents deployed in production. Success rates sharply decline as agent workflows become more complex, with overall reliability dropping below 50% after just thirteen sequential steps, even assuming 95% reliability for each individual step.
“It will still require several years for AI agents to effectively manage intricate, multi-step workflows,” stated Anushree Verma, a senior director analyst at Gartner. “Organizations will essentially need an ‘agentic brain’—a system capable of creating, executing, and overseeing these workflows.”
Security presents another hurdle. Prompt injection becomes more hazardous when agents are empowered to act, and agents necessitate governance akin to privileged user accounts, including role-based permissions, audit logging, and human oversight for critical actions.
Currently, agents are proving successful in clearly defined scenarios such as IT ticket prioritization and data extraction, but they encounter difficulties with cross-system workflows involving financial commitments or regulated decisions.
Dedication to open source
OpenAI’s choice to keep OpenClaw as an open-source project could alleviate some enterprise security worries by enabling organizations to inspect the code and customize implementations. However, as Gogia points out, open-source transparency alone does not fulfill all enterprise requirements concerning security controls, support models, and accountability.
Neither Altman nor Steinberger offered specific timelines for when agent capabilities might appear in OpenAI’s commercial products, although Altman suggested the technology would “rapidly become integral to our product offerings.”
Questions persist regarding how OpenClaw’s framework will integrate with OpenAI’s existing products and whether OpenAI plans to address the security issues identified in the open-source version.
