OpenAI Introduces gpt-realtime Speech-to-Speech Model
Today's curated insights on AI product management from X/Twitter across 60+ expert sources and YouTube channels.
OpenAI Introduces gpt-realtime Speech-to-Speech Model
From X
AI Product Launches & Updates
GPT-realtime speech-to-speech model and Realtime API updates: OpenAI Team @OpenAI introduced gpt-realtime—their best speech-to-speech model for developers—along with updates to the Realtime API, enabling real-time voice interactions.
Gemini Flash 2.5 image generation and editing: Jeff Dean @JeffDean shared an interview on the new image generation and editing capabilities in the latest Gemini Flash 2.5 model (aka nano banana), highlighting its improved speed and quality.
Nano Banana’s world-knowledge use cases: Jason Zhou @jasonzhou1993 highlighted how Nano Banana leverages world knowledge to generate visualizations of map locations, extending functionality beyond simple image editing.
AI Tools & Applications
Retrieval Augmented Generation course: Andrew Ng @DeepLearningAI announced a new course on RAG, teaching how to ground LLMs via retrieval to improve factuality, covering hallucination mitigation, prompt length tradeoffs, and compute costs.
Parallel agents for AI performance: Andrew Ng @DeepLearningAI explained how using parallel agents can scale AI speed and performance by solving tasks concurrently without making users wait.
GPT-5 API response issues: Claire Vo @clairevo reported encountering crazy typos, punctuation errors, and nonsense in the GPT-5 API responses, prompting escalation with @OpenAIDevs.
Product Management Insights & Strategies
Planning in “seasons” for dynamic roadmaps: Lenny Rachitsky @lennysan outlined Microsoft’s approach to manage frequent changes by organizing roadmap planning into “seasons” instead of fixed quarterly or annual cycles.
Tools of the Trade for discovery: Teresa Torres @ttorres shared the Tools of the Trade guide for continuous discovery and invited product teams to share their favorite discovery tools.
AI Industry Developments & News
OpenAI’s $30B Oracle data-center expansion: Andrew Ng @DeepLearningAI reported that OpenAI plans a 4.5 GW data-center build with Oracle under the “Stargate” program, involving a $30 billion annual deal following a 1.2 GW site in Texas.
Insights from Microsoft’s AI platform lead: Lenny Rachitsky @lennysan highlighted Asha Sharma’s role as AI platform product strategy lead at Microsoft, offering visibility into what works across 80,000 startups and enterprises.
From YouTube
How 80,000 companies build with AI: Products as organisms and the death of org charts | Asha Sharma
Lennys Podcast • August 28, 2025
Asha Sharma examines how AI is transforming products into continuously learning “organisms,” forecasts an “agentic society” where AI agents replace traditional org charts with task-based work charts, and describes the shift from GUIs to code-native, post-trained models.
Key Takeaways:
- Once a foundation model exceeds 30 billion parameters, Asha Sharma argues it’s more cost-effective to improve it via post-training loops (fine-tuning and reinforcement learning) than to invest further in pre-training.
- Microsoft’s AI platform already supports over 15,000 customers deploying millions of AI agents, signaling the rise of an “agentic society” where org charts become dynamic “work charts.”
- AI products are moving from static GUIs to code-native, composable interfaces—leveraging text streams, APIs, and agent collaborations—to enable seamless model integration and continuous evolution.