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, Perplexity AI launched a new interactive language-learning experience that embeds cards in the token stream. It’s available now on iOS and web, with Android support arriving soon.
In related news, Anthropic’s Claude introduced Skills—instruction packages that guide workflows across chat, code, and APIs. These Skills can be stacked to orchestrate tasks.
Another update from OpenAI: storyboards are now on the web for Pro subscribers, and video generation has been extended so that all users can produce up to 15-second clips, while Pro users can reach 25 seconds.
Moving to AI tools, Cognition rolled out SWE-grep and its mini variant, delivering agentic search at over 2,800 transactions per second and surfacing files up to 20 times faster thanks to its Fast Context subagent.
Additionally, Google’s AI Studio now offers a unified playground for chat, generative media, and live APIs across the Gemini suite.
Meanwhile, There’s an AI for It unveiled Bee, a wearable conversation summarizer that transforms speech into summaries, reminders, and reflections. It supports 40 languages, filters background noise, and runs for a week.
Separately, Fireship highlighted the Free Software Foundation’s Libriophone initiative, led by Rob Savois. It aims to replace proprietary firmware, drivers, and Google Play services on modern smartphones with fully free software. Earlier projects like Replicant struggled to support Wi-Fi, cameras, and GPS, limiting adoption.
Turning to PM insights, Andrew Ng emphasized that rigorous evaluation and error analysis at every step is the single biggest predictor of AI agent progress.
On the design front, Lenny Rachitsky spoke with Figma’s CEO about why design, craft, and quality form the new startup moat, lessons from the Adobe deal, and team motivation strategies.
In other best practices, Madhu Guru stressed that defining clear agent evaluation specs and constraining input-output spaces are core PM tasks to manage endless use cases and failure modes.
In industry developments, Google DeepMind announced a research partnership with Commonwealth Fusion Systems to accelerate fusion power using AI.
At the same time, Andrej Karpathy reported that his nanochat d32 model finished training in about 33 hours, achieving a CORE score of 0.31—above GPT-2’s 0.26—and boosting GSM8K performance from around 8% to 20%.
Another update on Windows: vision capabilities are now generally available globally, bringing us closer to an AI-driven operating system you can talk to and that sees your surroundings.
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!