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.
Cognition just closed a $1 billion Series D at a $26 billion valuation, reporting enterprise usage up more than tenfold and a $492 million run-rate. Their internal assistant, Devin, now authors 89 percent of pull requests and powers features like Devin Review and Auto-Triage. In related news, Anthropic’s Claude Marketplace added five new offerings—augmentcode, boltdotnew, coderabbitai, hebbia, and WeAreLegora—letting customers apply existing spend toward these live, Claude-powered tools. And Alibaba’s Qwen 3.5 hit a performance milestone of 580 transactions per second on its TokenSpeed engine for agentic workloads, thanks to FA4 optimization with NVIDIA and Lightseek.
On the product side, HubSpot rolled out a private-beta Agent CLI designed for both human operators and AI agents, promising a seamless “AX” or agentic experience alongside the traditional command-line interface. Meanwhile, LlamaIndex released LiteParse v2.0, rewritten in Rust for up to a hundred-times faster parsing with native support in Rust, JavaScript, TypeScript, and Python, plus a WASM package for browser or edge runtimes. xAI introduced grok-build-0.1 for high-speed, agentic coding intelligence in the Kilo IDE or CLI, available to SuperGrok and X Premium+ subscribers.
Shifting to AI-powered productivity tools, Peter Yang demonstrated Claude Code’s new /slides skill, which converts a rough outline into a fully animated HTML slide deck in minutes. It offers twelve prebuilt layouts, three visual templates, and an automated QA pipeline that reads style specs, asks clarifying questions, pulls in research, generates HTML, renders slides as images, identifies layout issues, and applies fixes—all in about three minutes versus an hour manually.
And if automation is your game, Claire Vo’s experience with Codex’s /goal command speaks volumes. She cleaned up 4,000 emails in four hours, while separate demo runs showed Codex eliminating thousands of Sentry errors in under six hours, whittling 3,900 messages down to 68 unread, and processing hundreds of project issues in less than an hour.
On the strategy front, Dharmesh Shah urged PMs to look beyond the frontier model race and instead build workflow systems around proprietary data for reliable, deterministic outcomes. Complementing that, Dan Shipper warned that AI speeds up change so fast that human judgment is crucial to detect shifting frames and reframe work. Udi Menkes suggested creating a personal SOUL.md style guide to load before every AI prompt, codifying your voice, values, and refusal criteria. Cost control also came up: Harrison Chase highlighted LangSmith’s LLM Gateway as essential for managing model spend at scale.
In industry headlines, OpenAI’s initial $250 million commitment to its new foundation will support measurement, transition assistance, and inclusive prosperity in the AI era. NVIDIA spotlighted HaoAI Lab’s FastVideo Dreamverse, cutting a five-second video generation workflow from 25 seconds on eight GPUs down to 4.2 seconds on a single GPU. They also unveiled Dynamo Snapshot, reducing cold-start inference on Kubernetes from minutes to under five seconds. And in infrastructure monitoring, Vercel’s AI-driven anomaly detection flagged a GitHub outage 16 minutes before the official status update.
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