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 top developments shaping the future of AI products.
OpenAI unveiled GPT-5.6 Luna, boosting health intelligence while costing 25 times less than GPT-5.5 at its highest reasoning setting. Microsoft has already integrated this model into Microsoft 365 Copilot, improving reasoning in office workflows.
On the product front, Cursor added side chats for parallel agent conversations, searchable transcripts, simpler project pickers, and new cloud agent hooks to streamline team collaboration.
In tools, Jason Zhou released Loopany, an open-source framework for designing and running self-improving agent loops with scaffolding, triggers, and templates. Viktor debuted as an AI “employee” in Slack and Teams, automating tasks across over 3,000 integrations and offering $100 in new-user credits. Meanwhile, Hugging Face rolled out HuggingNews, an AI-powered feed that curates top AI stories and will personalize recommendations via Hugging Face profiles.
Turning to product strategy, Lenny Rachitsky noted “the strongest product leaders are less visionaries and more curators,” highlighting the value of assembling teams and refining ideas. Garry Tan added that retention stems from “a real awesome product” rather than band-aid churn-prevention tactics.
In industry news, Meta released Muse Spark 1.1, an agentic model priced below competing offerings and boosting Meta’s stock by over 10 percent. OpenAI expanded its Bio Bug Bounty, doubling rewards to $50,000 and inviting security experts to find universal jailbreaks on frontier models. Google DeepMind launched a podcast exploring mechanistic interpretability, chain-of-thought monitoring, safety audits, and future research directions.
A demo showed Grok 4.5 powering an Orgo agent that spun up a cloud VM, built a one-page startup landing site in seconds, and generated ten startup ideas, video thumbnails, a market insights brief, and a cold-email sequence. The session leveraged connectors for email, phone, code editing, and video thumbnail creation.
In benchmarks, GPT-5.6 Soul—released June 26 after a 30-day government review and initially available to about 20 trusted partners—added adjustable Max Reasoning and an Ultra mode. It scored 91.9 percent on Terminal Bench 2.1, achieved 54 percent on the UC Berkeley–led Agent’s Last Exam at roughly one-third the cost of Claude Fable’s 45 percent, and led by 0.7 percent on Zapier’s Automation Bench. On the VIBE Code Bench, Soul delivered an 81 percent code completion score, while Muse Spark hit 72 percent at approximately 35× lower cost.
Sensor Tower data reveal that the Solid Starts baby food app earns $1 million per month. AI-native companies are storing meeting notes and support transcripts as Markdown to fuel daily prototyping loops that drive feature innovation and boost engagement. And Jeff Bezos’s two-pizza-team principle is evolving—many AI startups now run with just one engineer paired with a salesperson.
Hands-on tests by Peter Yang compared GPT-5.6 Soul and Claude Fable 5 across six real-world tasks—from building an interactive travel site and a 3D space shooter to automating video clipping, adding an app nutrition tab, delivering business advice, and cleaning up an AI OS. Soul ran about 50 percent cheaper per API call and excelled in browser automation; Claude Fable edged ahead on barrel-roll support, and under a ChatGPT Max subscription, Soul never hit rate limits.
Finally, Instagram explained that its feed recommender uses high-dimensional embeddings. A new prototype applies a language model to translate embedding clusters into human-readable topics—like “deep pour-over coffee”—and lets users view and adjust their algorithmic interests directly.
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