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.
In API news, OpenAI released GPT-5.5 and GPT-5.5 Pro to its public API. Cursor integrated GPT-5.5, hitting 72.8% on CursorBench with a 50% discount through May 2, and introduced a new /multitask command to run parallel asynchronous subagents on queued messages.
On the tools side, Cognition rolled out a GPT-5.5 agent preview in Devin for extended runtime and end-to-end issue detection. NotebookLM now auto-labels and categorizes sources to speed research. Google’s Gemini Deep Research API added collaborative planning so PMs can draft, refine, and approve research plans before execution. Meanwhile, Peter Yang used GPT-5.5 and Codex to build a live Star Fox clone and an F-Zero racing bot via prompt engineering, and compared ChatGPT Images 2 against Nano Banana 2 for visual asset creation.
For strategic insights, Lenny Rachitsky outlined ten key shifts for AI PMs. Shreyas Doshi urged teams to unlearn outdated assumptions as AI amplifies talent. Kevin Yien reminded us that today’s feature can become tomorrow’s bug, calling for constant reassessment. And Carl Vellotti’s “Team OS” knowledge base documented every decision to supercharge a junior PM—he’s now hosting a workshop on building your own AI-driven team operation system.
Industry headlines include Anthropic’s Project Deal, a live marketplace where Claude agents negotiate transactions on employees’ behalf; Meta’s expansion with tens of millions of AWS Graviton cores to power its AI and agentic experiences; NVIDIA’s day-zero metrics for DeepSeek-V4-Pro on Blackwell Ultra hardware; and Dharmesh Shah’s quip about teams up at 3 a.m. rethinking their LLM choices.
Moving to benchmarks, OpenAI reports GPT-5.5 scores about 6% below Opus 4.7 and 20% below Mythos Preview on the Agentic Coding Swebench Pro, but outperforms Mythos on Agentic Terminal Coding with 82.7%. NVIDIA’s DeepSeek V4 Pro supports up to a one-million-token context window via a 1.6-trillion-parameter mixture-of-experts model activating 49 billion parameters at inference.
In direct comparisons, GPT-5.5’s Codex output generated a fully functional F-Zero racer with boost mechanics and competitors, while Opus 4.7’s version ended abruptly. For a corgi café website, Opus delivered smooth hover-zoom effects; GPT-5.5 added unique button styling and smooth scrolling but showed some text overlap. On fitness advice, GPT-5.5 prioritized progressive strength training and high protein, versus percentile-based muscle assessments from Opus.
Claraveo tapped GPT-5.5 Pro in Codex to triage security CSVs, migrate two million chat logs in six hours with 98% edge-case coverage, and reverse-engineer a Doom Mini2 Bluetooth display into a CLI bitmap sender. Token pricing stands at $5 per million input and $30 per million output tokens for GPT-5.5, and $30 input/$180 output for the Pro tier.
Lastly, Anthropic’s product team slashed feature timelines from six months to as little as one day by shipping weekly research previews and running an “evergreen launch room” where engineers post ready features and marketing produces announcements within 24 hours.
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