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.
Thinking Machines introduced Inkling 2 Mini, a 26-billion-parameter looped transformer built for agentic work and sized to run locally on a laptop. The company reports a 20.1 AA Index score, signaling continued movement toward capable local agent models.
Google unveiled its Gemini 4 lineup: Neon for complex software, knowledge work, and scientific tasks; Hydrogen for faster coding and agent workflows; and Carbon for long-horizon research. Gemini 4 Neon is already demonstrating interactive 3D generation, including a live Mont Blanc ski simulator with custom shaders, real-time snow and avalanches, and no pre-made 3D models.
Neon is available through Gemini, AI Studio, and Antigravity for raw-code motion graphics, physics simulations, WebGPU games, app development, and multi-agent builds. Reported Antigravity coding performance rose from 12.8% to 39%. Google said Neon’s high-thinking mode is 32% cheaper than Carbon, while extra-high thinking uses 70% more compute for under two additional points.
On the agent front, Cognition highlighted Grant Sanderson’s Devin Bot setup for maintaining Manim, an open-source animation project. It’s an example of giving an agent continuing ownership of an operational area, rather than assigning isolated tasks.
Alexandr Wang announced that Muse can turn camera photos into polished e-commerce product shots. Guillermo Rauch shared an open-stack personal-shopper agent template, built around improved models, mature iMessage and WhatsApp channels, durability, and encryption.
For evaluation, Santiago Pino pointed to more than 1,200 reinforcement-learning environments based on real work across healthcare, life sciences, legal, and finance. The shift is toward training and evaluating models on domain-specific workflows, not generic benchmarks.
Claire Vo outlined a code-review progression: improve deterministic checks, apply risk review, then auto-merge low-risk changes. Amplitude reportedly reached 81% auto-merge while improving speed and quality. Udi Menkes cited Reuters Institute findings that agent launches are outpacing meaningful delegation, making the journey from “tried it” to completing a first real task a key adoption metric.
In go-to-market news, Dharmesh Shah argues that “agentic growth” is coming as buyers use personal agents to discover, compare, and select products. Product information, pricing, capabilities, and trust evidence need to be understandable not just to buyers, but to their agents. Shah also advised leaders to favor goal-seeking team members over people focused on internal visibility.
AI governance is becoming a platform differentiator. Vercel completed ISO/IEC 42001:2023 certification, covering how AI systems are built, evaluated, monitored, and governed.
In science, OpenAI is collaborating with PerturbAI Therapeutics and Scripps Research on AI-driven neurodegenerative disease research. Isomorphic Labs and Merck are collaborating to design a cancer drug using AI.
A proposed AI-assisted proof of the Riemann hypothesis is reported as a roughly 17-million-line Lean codebase. Claude reportedly helped near the finish line, while mathematicians still need to verify the formalized theorem matches the hypothesis. The Clay prize requires publication and two years of review; a separate, all-human proposed proof of the generalized version was also announced.
Finally, security teams are tracking three major exploits: malware delivered through 178 Optimizely-loaded sites, Java 17-and-later remote code execution via malformed Base64 input, and an Nginx HTTP/2 flow-control flaw enabling remote command execution. These incidents reinforce the need to treat third-party scripts, infrastructure dependencies, and AI product integrations as core attack-surface decisions.
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!