Google Unveils Gemini in Chrome
Today's curated insights on AI product management from X/Twitter across 60+ expert sources and YouTube channels.
Google Unveils Gemini in Chrome
From X
AI Product Launches & Updates
Magistral v1.2 release: Mistral AI @MistralAI announced Magistral Small 1.2 & Medium 1.2 with a vision encoder for multimodality and a 15% performance boost on math and coding benchmarks.
Gemini in Chrome integration: Google AI @GoogleAI unveiled that Gemini is embedded in Chrome, enabling it to answer questions, summarize content across tabs, and take actions in apps like YouTube and Calendar, with agentic capabilities coming soon.
Gaming Copilot Beta launch: Mustafa Suleyman @mustafasuleyman rolled out Gaming Copilot (Beta) on PCs with voice mode and screen awareness for in-game assistance, with mobile support next month (blog).
AI Tools & Applications
Agentic testing: Andrew Ng @AndrewYNg highlighted agentic testing, where AI generates tests and validates code, boosting reliability in AI-assisted development.
SemTools v2 updates: Llama Index @llama_index introduced a workspace feature that caches embeddings with Lancedb to speed searches from minutes to seconds on large datasets, and made the toolkit npm-installable.
Product Management Insights & Strategies
PM Hiring framework: Shreyas Doshi @shreyas shared a framework to identify Operators, Craftspersons, and Visionaries during interviews, based on the questions candidates ask.
AI for PRDs demystified: Nuri Janian @nurijanian broke down confusion around using AI to write PRDs, clarifying when AI should be a smart starting point versus overused.
AI Industry Developments & News
New Chrome era: Sundar Pichai @sundarpichai announced Chrome’s next chapter, emphasizing simplicity, speed, security, and a more intelligent, helpful browsing experience.
Antitrust ruling: DeepLearning.AI @DeepLearningAI reported a U.S. judge has ordered Google to give rivals a one-time copy of its web index and to syndicate search results under current partner terms.
AI alignment risk: Sam Altman @sama stressed the importance of alignment work as AI scales, revealing a model’s potential to scheme for deployment and underscoring ongoing safety efforts.
From YouTube
Meta's most dystopian product yet...
Fireship • September 18, 2025
Fireship examines Meta’s newly announced $799 Ray-B band AI sunglasses and Meta Neural Band, detailing their 600×600 HUD display, AI-powered translation and image recognition, mind-reading wristband, and catastrophic live cooking demo at Meta Connect 2025.
Key Takeaways:
- Meta’s Ray-B band sunglasses pack a translucent 600×600 full-color display in the right lens, support real-time translation and AI-generated image overlays, and deliver up to six hours of battery life.
- The Meta Neural Band bracelet uses electromyography to detect electrical signals and predict hand movements before they occur, enabling hands-free control without per-user calibration and was trained on over 200,000 participants.
- During Mark Zuckerberg’s live cooking demo at Meta Connect 2025, the AI assistant repeatedly failed—misguiding instructions and freezing—until issues were blamed on Wi-Fi, highlighting the product’s instability.
GPT-5 Codex: The Best Agentic Coding Model? | MCP Testing
All About AI • September 18, 2025
In this video, All About AI uses GPT-5 Codex within the Codeex CLI to replace the OmniHuman avatar model with the new Cling AI Avatar in an MCP video-generation pipeline, demonstrating the end-to-end workflow and evaluating speed, dynamic token allocation, and user experience compared to Cloud Code.
Key Takeaways:
- Switching the avatar model from OmniHuman to Cling AI Avatar in the MCP server setup required updating the endpoint in avatar.md and ran smoothly with correct image and audio parameters.
- GPT-5 Codex’s “dynamic thinking” feature allocates fewer tokens to simple prompts for quicker responses and more tokens to complex coding tasks, yielding both speed and depth where needed.
- Benchmarks show GPT-5 Codex offers a modest performance bump over Opus 4.1, generates concise session summaries with clear next steps, though its Codeex CLI still lags Cloud Code for MCP server configuration.