Stanford-Princeton Team Open-Sources LabClaw
AI Product Management Certification
Today's top 14 insights for PM Builders, ranked by relevance from X, Blogs, LinkedIn, and YouTube.
Stanford-Princeton Team Open-Sources LabClaw
#1 đť•Ź
AI4Science Catalyst has open-sourced LabClaw, the Skill Operating Layer for LabOS by the Stanford–Princeton team, letting you turn any OpenClaw agent into a full AI Co-Scientist with a single command. Try the demo at labclaw-ai.github.io.
#2 đť•Ź
PM Diego Granados set up a nightly OpenClaw Use Case Research cron that auto-opens a browser to openclaw.ai plus Reddit and X, runs search extensions, and follows at least one resulting link to gather richer context.
#3 đť•Ź
Tal Raviv discovered that while building a desktop app with Claude Code, its Familiar skill automatically used the markdown “source image” field to interpret and act on his UI screenshots—even though image support isn’t documented.
#4 đť•Ź
Sebastian Raschka launched a new LLM Architecture Gallery, consolidating architecture figures of major large language models into one centralized reference.
#5 đź“ť Simon Willison
What is agentic engineering? - Defines 'agentic engineering' as the practice of developing software with the assistance of coding agents—systems that can both write and execute code—and gives examples and context. The piece situates the term within LLM-driven tooling and points to further reading.
#6 in
Udi Menkes has been automating his PM work in Claude Code and, echoing Ramp’s CPO Geoff Charles, now treats the prompt as the spec and submits PRs directly to engineering.
#7 in
Peter Yang unveils a new episode with Ramp CPO Geoff where he breaks down an AI-native playbook—using Claude Code, custom AI agents for research, data & coding, plus an L0-L3 framework to get every employee shipping production code.
#8 ▶️
Inside Ramp, the $32B Company Where AI Agents Run Everything | Geoff Charles
Peter Yang
Ramp uses Claude-powered Cloud Code and the Inspect AI agent to convert PM prompts into production-ready front-end and back-end features complete with pull requests in under five minutes.
- 50% of Ramp’s code is built by AI (up from 30% in December), with a projection to reach 80% by March.
- The Voice of the Customer agent processed 90 days of support tickets and chat logs in eight minutes, surfacing issues such as purchase order management, approval flow routing, exports, and currency constraints.
- Inspect generated a full accounts payable report—overdue bills for 0–30 and 30–90 days plus total amount outstanding—by interpreting a five-line prompt and accessing the codebase and design system in under five minutes.
#9 đť•Ź
Andrej Karpathy clarifies that his LLM-based “exposure” score simply measures how digital a job is, not its real-world AI displacement risk, and warns that true outcomes depend on factors like demand elasticity—not his visualization tool.
#10 in
Birkan Icacan highlights Ramp CPO Geoff Charles’s demo of a Slack-embedded Voice of the Customer agent that turns eight days of research into eight minutes of insights.
#11 đť•Ź
There's An AI For That: @AI4S_Catalyst just launched LabClaw, a skill library that empowers AI co-scientists to autonomously execute real laboratory workflows.
#12 đť•Ź
PM Diego Granados pulls Discord announcements into his Obsidian Vault for easy note access and has built in a fully logged “kill switch” to instantly yank the plug if anything goes off-script.
#13 đť•Ź
Sebastian Raschka published YAML-formatted metadata for his llm-architecture-gallery on GitHub (https://github.com/rasbt/llm-architecture-gallery), making it easy to browse and tinker with various model designs.
#14 in
Tal Raviv built with Claude Code and the Familiar skill on a desktop app and was amazed it automatically pulled in and processed images from the markdown “source image” field—even though the skill never mentions image support.