GPT-5-Codex is now available in Cursor

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. Starting with product launches: Cursor AI has integrated GPT-5-Codex into its coding assistant, giving developers access to the latest language model for more accurate code generation and real-time suggestions. In related news, Alibaba’s Qwen team open-sourced its flagship Qwen3-VL-235B-A22B vision-language model in both Instruct and Thinking versions—claiming leading scores on key benchmarks. And on the same front, Alibaba also rolled out Qwen3-Max, which boosts coding and agentic capabilities with a new Max-Instruct mode that rivals top models on SWE-Bench and Tau2-Bench. On the tool side, Claude demonstrated a design-to-code workflow: Claude Code now plugs into Figma via MCP, translating mockups into production-ready code at the component level. Separately, LangChain announced an upcoming webinar featuring experts from LangChain and ManusAI, covering proven patterns and blueprints for production-grade GenAI agents. Additionally, LangChain launched Composite Evaluators in LangSmith, enabling weighted aggregation of multiple evaluation scores into a single performance metric—making it easier to compare models across diverse criteria. On the product front, Lenny Rachitsky shared a two-part series on how he uses each tool in his Product Pass, offering actionable tips for PMs to integrate these resources into their workflows. Another development comes from Nurijanian, who highlighted research showing that excessive context can actually harm AI model performance, and distilled 12 practical rules to prevent context overload. Meanwhile, Teresa Torres discussed key strategies on her AI Evals & Discovery podcast—covering how to define, measure, and operationalize AI evaluation metrics to validate product performance effectively. In industry news, OpenAI announced five new Stargate compute sites built in partnership with Oracle and SoftBank, accelerating its 10-gigawatt infrastructure rollout ahead of schedule. On a different front, Demis Hassabis revealed enhancements to the Frontier Safety Framework, expanding risk domains and refining assessment protocols for advanced AI systems. Finally, Google Research presented a novel approach to time-series foundation models: by continuing pre-training, these models can perform few-shot forecasting that matches supervised fine-tuning—without extra training steps. 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!