Welcome to GenAI PM Daily, your AI product management news. I'm your host, and here are today’s top developments shaping the AI product management landscape.
Cognition launched Devin Desktop, a Windsurf-based console for managing local and cloud agent fleets with full integrations and Agent Client Protocol support.
Meanwhile, Mustafa Suleyman released seven MAI models with 35B MoE parameters, 256K context windows and 97% AIME 2025 performance, including image and code options.
OpenAI rolled out a public equity investing plugin for Codex that enables model-driven market data analysis.
In developer tools, Philipp Schmid introduced GEPA, a Python tool that optimizes prompts for any string-to-string agent CLI.
Santiago Valdarrama unveiled Bigset, an open-source orchestrator that spawns parallel sub-agents to fetch structured datasets with scheduling and self-hosting.
Clement Delangue showcased Arcee, a U.S. open-source model on Hugging Face offering a smaller, lower-cost alternative to frontier models.
In strategic thinking, Garry Tan advised using dynamic routing across commoditized models to balance cost and performance, and writing Markdown-based tasks that let agents iteratively craft tools in a continuous kaizen cycle.
Peter Yang noted narrow vertical apps face monetization challenges as AI-native agents deliver flexible, context-rich solutions and enterprises prefer multi-function platforms.
Anthropic’s Project Glasswing expanded Claude Mythos Preview to around 150 organizations in more than 15 countries.
Anthropic praised the White House’s AI Executive Order on innovation and security and pledged to support it.
Separately, Mustafa Suleyman announced a partnership with Mayo Clinic to co-develop a frontier AI model for healthcare with global impact.
Over in telephony, Saperly launched an API-first platform letting AI agents provision and manage real phone numbers in minutes versus days, offering telephony, SMS, transcription, billing and webhooks, with 90% of users in business workflows.
Greg Isenberg outlined six pillars of the agent-first internet—identity, tools, inbox, memory, wallet and receipts—shown in AgentMail’s AI inbox API and Stripe’s agent wallet, urging sites to add a /agents endpoint with structured docs and SDKs.
For builders, Andrew Ng demonstrated prompting ChatGPT, Gemini and Claude to generate full HTML apps in minutes—a birthday card tool with custom inputs and autofill, and a ping-pong game with paddle controls, difficulty levels and live scoring.
Loopany used open-source agent skills, a cron-driven memory layer and HubSpot’s AEO Grader to build closed-loop SEO and ad optimization; one autonomous loop tested multiple ad formats and generated 243 leads on a $1,500 budget.
Finally, Moda showcased an AI agent inside a Figma-like canvas that programmatically places text and shapes to produce editable, pixel-perfect PowerPoint decks from a three-sentence prompt; the YC-backed startup has raised $23 million and reached double-digit million ARR.
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