OpenAI Previews Upcoming AI Models, Features

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. OpenAI CEO Sam Altman announced a slate of new models, products, and features slated to launch over the next couple of months, though he cautioned teams to expect potential hiccups and capacity crunches. Meanwhile, DeepMind’s Demis Hassabis revealed that Gemini 2.5 Deep Think is now delivering state-of-the-art performance across some of the toughest AI benchmarks. On a different front, Claire Vo introduced the chatprd MCP, which is now live and seamlessly integrates with Cursor, Windsurf, Claude AI and VS Code—giving product teams the ability to tap multiple tools within a single chat interface. In new tools coverage, Sebastian Raschka demonstrated a PyTorch-based Mixture-of-Experts setup with Qwen3 Coder Flash (30B-A3B), featuring 128 experts with eight active per token, all running on a single A100 GPU in a Jupyter notebook. Additionally, LangChainAI released a comprehensive retrieval-augmented generation pipeline for internal documentation, complete with multi-LLM support and ChromaDB integration for both experimental notebooks and production environments. Alongside that, they rolled out an open-source multilingual audio conversation tool that transforms text, images, websites and videos into dynamic, multi-speaker audio dialogues via local LLM deployment. In strategic insights for product leaders, Lenny Rachitsky urged founders to challenge their natural superpowers—encouraging self-awareness instead of defaulting to familiar problem-solving strengths. He also recounted how Brian Taylor’s biggest product failure ultimately evolved into Google Maps. From a career perspective, Aakash Gupta laid out a four-step roadmap to break into AI product management: reverse engineer top AI products, conduct teardown case studies, build a side project, and focus on synthesis. In broader industry news, Andrej Karpathy noted that 2024 has been dominated by chat models, while 2025 is shaping up to be the year of code-centric AI releases. Clement Delangue posed the question of whether open-weights infrastructure has now matched or even surpassed proprietary API ecosystems. Meanwhile, DeepLearningAI outlined the EU’s voluntary General-Purpose AI Code of Practice under the AI Act, which calls on developers to document data sources and log model risks throughout development. In on-the-ground experimentation, the video titled Veo 3 Advanced Prompting showcased three autonomous prompting agents—JSON, XML and enhanced NLP—within Veo V3. It compared their outputs across an exploding-box IKEA-style reveal, a rainy street interview, and a Minecraft obsidian ASMR slice. In the IKEA test, the NLP prompt hit precise timing but glitched in the final seconds, while XML delivered a more powerful explosion and cleaner audio. During the eight-second street interview, XML again led with the most natural handheld footage and authentic rain ambience. And in the ASMR obsidian slicing scenario, the JSON prompt produced the cleanest slice animation, outperforming XML’s awkward diagonal cut. 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!