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.
Leading the product news, OpenAI expanded ChatGPT for Healthcare with stronger clinical responses, more capable computer-use agents, and more than 20 plugins for prior authorizations, referrals, discharge planning, and patient outreach.
Hugging Face added direct OpenRouter integration, with one token and endpoint, live Hub model statistics, and a redesigned inference widget. Google DeepMind is providing 10,000 scientists free Gemini 4 Carbon access through its expanded Antigravity program.
On the tools front, Perplexity released Monitor, an API that tracks events, topics, and webpage changes, then fires webhooks for competitive intelligence or research. Its Teammate on Fable 5.2 built a browser-based open-world driving game: roughly 63,000 lines of JavaScript, its own renderer, and generated assets. Bot now lets users choose Anthropic models using their own API keys.
At HubSpot, Dharmesh Shah outlined agents as governed users, with provisioned identities, role-based permissions, and audit logs. More CRM capabilities are being exposed through APIs and MCP, connecting assistants to customer data across surfaces. Guillermo Rauch says agent products need evals alongside user research: repeatable tests of task completion. He calls traces—prompts, intended actions, tool use, latency, and outcomes—a core asset for improving behavior.
Greg Isenberg’s playbook says build an organizational brain before scaling agents: capture decision principles, customer conversations, approved facts, and workflows; turn jobs into reusable skills; trigger them appropriately; measure cost and value; then feed results back. Peter Yang recommends giving Claude product context, researching familiar patterns, starting with wireframes, generating alternatives, and documenting recurring visual biases to counter them. Garry Tan proposed a Replication Appendix containing runnable queries, signal definitions, assumptions, rebuild instructions, and limitations for agent-assisted reports.
Harrison Chase said advancing models are shifting from executing tasks to pursuing goals, putting agent safeguards, interaction patterns, and delegation boundaries at the center of product design.
In industry news, OpenAI introduced embedded assessments with Apollo Research, METR, Transluce, and Goodfire, giving outside experts deep access to assess internal models and safeguards. Isomorphic Labs raised $5 billion to accelerate AI drug design. Databricks CEO Ali Ghodsi reported enterprise gateway data showing open-source models exceed 30 percent of tokens, while most enterprises use multiple providers.
Finally, a Microsoft network capacity upgrade exposed a latent secure-connection bug, disrupting Exchange Online, Outlook, and other Microsoft 365 services for roughly 36 hours. Mailboxes stopped opening at 11:17 a.m. Eastern Thursday; Microsoft acknowledged the investigation about 90 minutes later, while reports arrived from 74 countries. Failed health checks removed servers from load balancing, concentrating traffic on remaining systems; restarted servers failed again as clients reconnected. Microsoft rolled back a network configuration change. Microsoft and the Homeland Security Secretary said the FAA’s roughly five-hour departure halt was unrelated.
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!