OpenAI Hires OpenClaw’s Founder to Lead Personal Agents
Today's top 11 insights for PM Builders, ranked by relevance from X, YouTube, Blogs, and LinkedIn.
OpenAI Hires OpenClaw’s Founder to Lead Personal Agents
#1 𝕏
Sam Altman announced Peter Steinberger is joining OpenAI to spearhead next-generation multi-agent personal assistants, and that OpenClaw will be open-sourced under a foundation to support this future.
#2 ▶️
How to Run OpenCode Inside an Autonomous Claude Code AI Agent
All About AI
Uses an autonomous Claude Code agent on a Mac Mini to invoke the OpenCode CLI via OpenRouter on four models (GLM5, Minimax 2.5, Gemini 3 Pro, Opus 4.6) in parallel to generate HTML demos of a retro space game, convert them with Remotion into a grid-style MP4 video, and draft a post on X.
- Executed “open code run --model openrouter GLM5 'Should I walk or drive to the car wash? It’s 50 m away'” via Cloud Code CLI, receiving “you should walk to the car wash,” and then ran “open code run --model openrouter Gemini-3-Pro …” obtaining “drive. You can’t wash the car if you leave it behind.”
- Created a Cloud Code skill file open code test skill.md to launch four OpenRouter models (GLM5, Minimax-2.5, Gemini-3-Pro, Opus-4.6) in parallel on the prompt “create a full screen animated retro arcade space battle scene,” saving outputs as llm-test/game-
.html. - Applied Remotion to merge the four HTML files into a 15-second grid-style MP4 video (retro-space-benchmark.mp4) and composed a draft X post with the clip and the caption “Gave four LLMs the same prompt. Build a retro arcade space game demo in a single HTML 5. Here’s how GLM 5, Minimax 2.5, Gemini 3 Pro, and Claude Opus did side by side.”
#3 ▶️
Full Tutorial: The Most Underrated AI Agent for Coding and Product Work | Eno Reyes (Factory)
Peter Yang
Uses Factory’s Droid agent via the Ghosty CLI in high-autonomy spec mode with Opus 4.5 for planning and GPT-5.2 for execution to build and QA a React-based speed-reading web app using Chrome DevTools for automated screenshots, linting and type-checking.
- Spec mode (Shift+Tab) prompted for input sources (set to “all of the above”) and reading enhancements (chunk mode, party mode) then generated an editable spec document in VS Code before running the plan.
- Droid autonomously opened a browser, took screenshots via Chrome DevTools, ran continuous linting and type-checking, performed React hot reload, and validated console errors on the speed-reading app displaying two-word chunks.
- Droid supports model switching mid-session, using Opus 4.5 for high-level planning intelligence and GPT-5.2 codeex for diligent code execution and self-validation.
#4 𝕏
Jason Zhou walks through configuring webMCP via HTML attributes or a React setup to instantly make websites agent-ready. He invites you to @aibuilderclub_ for a deeper breakdown and live walkthrough in his upcoming weekly call.
Also covered by: @AI Jason
#5 📝 Anthropic Engineering
Quantifying infrastructure noise in agentic coding evals - An analysis showing that infrastructure configuration can materially change agentic coding benchmark results; differences from infrastructure can exceed leaderboard gaps between top models.
#6 📝 PromptLayer Blog
How Do Teams Identify Failure Cases in Production LLM Systems? - Explains that LLM failures are often subtle, context-dependent, and non-deterministic, making them hard to detect with traditional tooling. The piece draws on PromptLayer's experience to show common blind spots teams face and suggests approaches for surfacing these failure modes in production.
#7 📝 PromptLayer Blog
How Large Organizations and Enterprises Standardize LLM Benchmarks - Covers the challenge large organizations face when trying to evaluate LLMs consistently and meaningfully as models move into critical production roles. The article outlines the need for standardized, comparable benchmarks and shares PromptLayer's perspective on practical evaluation practices.
#8 in
Peter Yang launched the OpenClaw bot and was hired by OpenAI just three months later. He’s published quick tutorials—20-minute setup, 30-minute mastery with five real use cases + memory—and interviews on its personal and business applications.
#9 𝕏
Sebastian Raschka released Chapter 07 of his “Reasoning from Scratch” series, showcasing an enhanced from-scratch GRPO implementation. It adds clipped policy ratios, a KL divergence term, formatted rewards and other performance tweaks in the ch07_main.ipynb notebook.
#10 𝕏
Jason Zhou introduced WebMCP, a new Chrome 146 API that lets websites dynamically load and communicate with agents to perform page-specific actions, and shared steps to try it via Chrome Beta 146 and the WebMCP debugger tool.
Also covered by: @AI Jason
#11 in
Dharmesh Shah announces that Peter Steinberger, creator of the open-source personal agent OpenClaw (originally ClawdBot), is joining OpenAI. Sam Altman also revealed the project will spin out into its own foundation.