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.
Muse is broadening its agent surface with connectors, a connector platform, phone-call beta, Mac support and computer use, Canadian availability, small-business features, and gadgets. Claude Code Desktop is also iterating in public, with Anthropic seeking feedback on remaining product gaps.
Perplexity is opening vertical AI apps inside Computer, including image geolocation using 3D and satellite views. Its Decisions API powered an agent that cleared Pokémon FireRed’s Elite Four and Champion in 137 live calls, with 592-millisecond median latency, 96.4% of responses under one second, and roughly 2.8 cents in inference cost.
On agent economics, Harrison Chase highlighted three levers for reducing coding-agent spend: granular cost visibility, per-user caps through an LLM gateway, and model routing based on task economics.
The interface is shifting toward agent-first experiences. Thibault Sottiaux, who leads ChatGPT and Codex, expects model pickers to fade, while loops and graphs may be transitional patterns as agents take actions across the web. Udi Menkes argues every product now has two customers: the human buyer and their agent. Positioning, pricing, eligibility, claims, evidence, and the next permitted action must be legible to both.
Granola’s Sam Stephenson is focused on daily value, not merely shipping version one. Its most-used “What did I miss?” recipe summarizes missed meeting context, while automatic notes and pre-meeting briefs turn meeting history into reports and drafts. Internally, Granola’s Nacho agent connects Slack, product analytics, logs, and code to investigate incidents, run SQL, and create Cursor agents that can write pull requests.
Kath Korevec demonstrated ChatGPT Sites and Codex for permission-aware internal tools. An incident dashboard connects Slack, Notion, and calendars to show team-specific timelines, responders, runbooks, and tools. Sites supports around 60 integrations, plus D1 data, R2 storage, deployment, co-editing, and MCP plugin hosting. Her “heavy rotation” music site reviews Reddit playlists weekly using computer use.
OpenAI’s Dots are always-on personal agents that learn preferences across devices. Multiple specialist Dots are planned, with additional guardrails and separate hardware. OpenAI also plans shared economics for successful ChatGPT plugins, based on retention, usage, and quality.
In market news, Andrej Karpathy says mainstream AI awareness may have arrived in under a year. Yann LeCun expects cheap distillation to push frontier capabilities toward free or open models. Security pressure is rising: GLM-5.3 reached 12% on ExploitBench, near Claude Mythos at 14%, and one Chrome exploit reportedly cost $20.40 in tokens to find.
Greg Isenberg says cheap building shifts differentiation toward a memorable product thesis and daily value. Guillermo Rauch says agentic development could reshape technology choices, including language and platform investments. He also predicts security will become a larger core function as AI strengthens adversaries.
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