Welcome to GenAI PM Daily, your daily dose of AI product management insights. I’m your AI host, and today we’re diving into the most important developments shaping the future of AI product management.
On the product front, Demis Hassabis at Google DeepMind released the new Gemini 3.1 Flash-Lite model—“small but mighty” with incredible speed and cost-efficiency. Meanwhile, NotebookLM introduced Cinematic Video Overviews for Ultra-tier users, enabling polished video scripts complete with visual direction and critique loops. Also, DeepLearning.AI teamed up with Google to launch a JAX short course, guiding learners through building and training a 20-million-parameter MiniGPT-style model, covering architecture, preprocessing, training loops, checkpointing and a chat interface demo.
In tools news, Philipp Schmid published a skill for the Gemini Interaction API, installable via Vercel and Context7 CLIs to power advanced agentic applications. Additionally, Harrison Chase rolled out open-source skills for LangChain, LangGraph and DeepAgents, simplifying the development of agent capabilities across frameworks. Separately, LlamaIndex unveiled LlamaSplit, a document parser that breaks complex files into customizable categories and interactive sections through an intuitive user interface.
In strategy news, Peter Yang’s deep dive examines how companies like Linear, Ramp and FactoryAI onboard AI agents as virtual team members—assigning tasks, integrating reusable skills and slotting them into workflows. On a different front, Guillermo Rauch highlighted that “skills” are becoming the new onboarding experience for AI products, marking a paradigm shift in interaction. Another key development comes from Santiago Pino, who urged PMs to adopt an agent-first product strategy by embedding skills for agents like Claude, Codex and Cursor with clear navigation guidance, best practices and anti-patterns to avoid. And Adhyayan Rathi recommends storing product overviews, user personas, positioning and constraints as markdown context files in Claude Code, so they auto-load each session and keep outputs consistently grounded.
On the industry front, xAI outlined a sustainable infrastructure plan to power its supercomputers, featuring a 1.2-gigawatt power plant, the world’s largest Megapack installation, new substations, a 4.7-billion-gallon water recycling facility and thousands of jobs in Memphis. In related advances, Google Research introduced a method that teaches LLMs to think like Bayesians by mimicking optimal probabilistic inference, improving prediction updates and cross-domain generalization. Meanwhile, NVIDIA AI showcased micro data centers that tap underused substations for low-latency AI inference without overhauling the power grid. Additionally on LinkedIn, Guillermo Rauch unveiled “v0 Max Fast,” a premium generative service powered by Fast Opus for front-end and design outputs with remarkable speed, quality and consistency. Also, Dharmesh Shah predicts the rise of hybrid human–AI teams, sharing early sales prospecting experiments where AI assistants automate routine tasks and signal a broader shift PMs will need to design and manage.
Turning to video insights, Greg Isenberg outlined a 30-step AI-powered SaaS playbook using tools like ideabrowser.com, Manis, Claude Code and ChatGPT to uncover sub-niche markets, automate scroll-stopping content and shift from per-seat subscriptions to $200-per-task outcome pricing. And another tutorial demonstrated how to build and train a GPT-2-style LLM with 20 million parameters in JAX—leveraging automatic differentiation, just-in-time compilation and distributed compute—and then serve it via a graphical chat interface.
That’s a wrap on today’s GenAI PM Daily. Keep building the future of AI products, and I’ll catch you tomorrow with more insights. Until then, stay curious!
GenAI PM