OpenAI Hires OpenClaw Creator in Agent Push

Why it matters

- Before creating the agent framework OpenClaw, Peter Steinberger founded and built PSPDFKit, a widely used SDK for PDF rendering and annotation on mobile and web platforms. His background is in systems-level programming and developer tools, having been an active open-source contributor in the iOS community for years. - OpenClaw, initially named Clawdbot, became one of the fastest-growing open-source projects, amassing over 145,000 stars on GitHub shortly after its launch in November 2025. It functions as a self-hosted gateway that connects large language models to messaging apps like Telegram and Slack, allowing them to execute shell commands, control browsers, and manage local files. - The project's viral growth was fueled by its open-source nature and the popularity of Moltbook, a social network created for AI agents to interact with each other. This highlighted a significant developer interest in moving beyond conversational AI to autonomous, task-driven agents. - OpenAI's new GPT-5.2 model, which now fully powers ChatGPT, is architecturally designed for agentic workflows. It features a dynamic tier routing system to match compute resources to query complexity and an API endpoint that compacts conversational history to manage long-running tasks beyond its 400,000 token context window. - The move to hire Steinberger is part of a broader industry race to build agentic systems. Meta has also been actively developing its own AI agents, including an internal platform called "WhatsCode" for code modification and a reported upcoming model named "Avocado". - For ML Engineers, the shift towards agents introduces new operational challenges, extending MLOps to LLMOps. This requires managing new components like prompt templates, vector databases, and agent tools as first-class citizens in the production lifecycle. - Enterprise search competitors like Glean and Hebbia are already incorporating agent-like functionalities. Glean utilizes a knowledge graph to power an "AI Assistant" and "Glean Agents" for proactive tasks, while Hebbia's "Matrix" product uses AI to extract and structure information from private documents into a spreadsheet format. - The underlying infrastructure for serving these models continues to evolve, with frameworks like vLLM and TensorRT-LLM offering trade-offs between flexibility and performance. vLLM is often favored for its rapid iteration and handling of heterogeneous workloads, while TensorRT-LLM provides maximum throughput on NVIDIA hardware for stable, high-QPS services.

Key numbers

• Concurrently, OpenAI retired GPT-4o, moving ChatGPT fully to its new GPT-5.2 model.
• OpenClaw, initially named Clawdbot, became one of the fastest-growing open-source projects, amassing over 145,000 stars on GitHub shortly after its launch in November 2025.
• OpenAI's new GPT-5.2 model, which now fully powers ChatGPT, is architecturally designed for agentic workflows.
• It features a dynamic tier routing system to match compute resources to query complexity and an API endpoint that compacts conversational history to manage long-running tasks beyond its 400,000 token context window.

What happens next

• OpenClaw, initially named Clawdbot, became one of the fastest-growing open-source projects, amassing over 145,000 stars on GitHub shortly after its launch in November 2025.

Sources

• OpenAI hired
• will remain
• OpenAI retired
• Before creating the
• OpenClaw, initially
• It functions as a self-hosted
• This highlighted a significant
• OpenAI's new GPT-5
• It features a dynamic
• Meta has also been actively
• For ML Engineers, the
• This requires managing
• Glean utilizes a knowledge
• , while Hebbia's "Matrix"
• The underlying infrastructure
• vLLM is often favored

Quick answers