OpenAI Hires OpenClaw Agent Creator
OpenAI has hired Peter Steinberg, the developer of the open-source agent OpenClaw, to advance its robotics and agentic AI efforts. The hiring suggests OpenAI is intensifying its focus on embodied intelligence and the application of foundation models for physical manipulation and multi-agent control.
- OpenClaw, formerly Moltbot, is a self-hosted, open-source AI agent that gained over 160,000 GitHub stars shortly after its launch. It allows users to connect large language models to personal messaging apps like WhatsApp and Telegram to autonomously execute tasks such as shell commands, browser automation, and file operations. - Before joining OpenAI, Peter Steinberg was in acquisition talks with both Meta and OpenAI, with Mark Zuckerberg and Sam Altman personally involved. Steinberg has stipulated that OpenClaw will transition to an independent foundation to remain open-source. - This hiring signals a renewed push into robotics for OpenAI, which had previously disbanded its robotics team in 2021 to focus on the research that led to models like GPT-3. The company's earlier robotics work included Dactyl, a system that trained a robotic hand to solve a Rubik's Cube. - OpenAI's current robotics strategy heavily involves partnerships, most notably with humanoid robot company Figure. This collaboration allows the Figure 01 robot to use OpenAI's vision-language models to hold full conversations and translate them into learned, autonomous actions. - The underlying technology, often called a "robotics foundation model," aims to create general-purpose models pre-trained on diverse datasets, moving away from systems designed for a single, specific task. These models integrate multimodal data from sensors like cameras and lidar to generalize across a wide range of robotic scenarios. - The concept of "agentic AI," as seen in OpenClaw, is a key focus for applying foundation models to robotics, where an AI can autonomously plan and execute a series of actions to achieve a complex goal. This extends to multi-agent systems, where multiple robots might coordinate to perform tasks in environments like a warehouse or factory floor.
