OpenAI Rolls Out ChatGPT Apps
Today's curated insights on AI product management from X/Twitter across 60+ expert sources and YouTube channels.
OpenAI Rolls Out ChatGPT Apps
From X
AI Product Launches & Updates
Apps in ChatGPT Rollout: OpenAI @OpenAI shared that Apps in ChatGPT are starting to roll out today to Free, Go, Plus, and Pro users outside of the EU, with more pilot partner apps and availability coming soon.
Claude Sonnet 4.5 Release: Anthropic AI @AnthropicAI announced that Claude Sonnet 4.5 is released and open-sourced their alignment audit tool for testing behaviors like sycophancy and deception.
Copilot Memory Update: Mustafa Suleyman @mustafasuleyman highlighted that Copilot memory now supports both forgetting and remembering directions, with users able to check and edit memories under settings.
AI Tools & Applications
Comet Agent Prompting: Arav Srinivas @AravSrinivas shared a video explaining how to make good use of Comet for agent prompts.
Automated Security Patches with CodeMender: Google DeepMind @GoogleDeepMind introduced CodeMender, an AI agent using Gemini Deep Think to automatically patch critical software vulnerabilities.
Product Management Insights & Strategies
Growth Lessons from Duolingo & Grammarly: Lenny Rachitsky @lennysan shared insights from his interview with Albert Cheng on what good retention looks like for consumer subscription apps.
Building a PM Second Brain: Claire @clairevo highlighted how Amir at Monday.com created a second brain using ClaudeAI and ChatGPT to handle context switching, feedback analysis, and writing improvement.
Employee Motivation Framework: Dharmesh @dharmesh proposed a v2 model for why people join or stay at a company: MERCENARY (money), MISSIONARY (purpose), and MATES (community).
AI Industry Developments & News
AI Compute Partnership: Sam Altman @sama announced a partnership with AMD to use their chips to serve OpenAI users, complementing existing NVIDIA collaborations.
US AI Infrastructure Investment: Sundar Pichai @sundarpichai shared that Google is investing $4 billion in cloud and AI infrastructure in Arkansas through 2027, including a new data center in West Memphis.
ChatGPT Usage & SDK: Kevin Weil @kevinweil revealed that ChatGPT has 800 million weekly active users and the API handles over 6 billion tokens per minute, and introduced the upcoming Apps SDK for third-party integrations.
From YouTube
I Watched Dan Koe Break Down His AI Workflow OMG
Greg Isenberg • October 06, 2025
Greg Isenberg sits down with Dan Koe to unpack his full AI-powered content engine—from weekly newsletter ideation and repurposing validated tweet ideas into multi-platform posts and YouTube videos, to building custom LLM prompts that distill research, generate titles, and produce viral social media variations.
Key Takeaways:
- Dan Koe’s core process centers on a weekly newsletter that fuels two-to-three daily social posts—written first on X—and a YouTube video, allowing one validated idea to spread across Threads, Instagram, LinkedIn, TikTok and Shorts.
- Koe validates ideas by testing tweets on X, curates angles from niche YouTube’s most popular videos (without copying), and refines structures with a swipe file and tools like the SuperX extension.
- He uses multi-phase LLM prompts in ChatGPT, Gemini and Claude—first to pull context, summaries and psychological patterns from top-performing content, then to generate 20+ YouTube titles, deep post building blocks, and tweet variations in his own voice.
“I’m incapable of doing my job without AI”: How this PM uses Claude + ChatGPT as his second brain
How I AI Podcast • October 06, 2025
Monday.com product manager Amir Klein demonstrates how he builds per-project Claude and ChatGPT “second brains”—uploading decks, website PDFs and scraped Reddit data—then crafts custom GPTs to rewrite his Slack updates concisely and employs ChatGPT-4 Voice Mode for realistic mock interviews, dramatically boosting his PM efficiency.
Key Takeaways:
- Klein organizes each initiative in Claude and ChatGPT as a separate “brain” with over 20 files and detailed instructions per project, helping him maintain context across back-to-back PM tasks.
- He prompted Claude to generate a step-by-step Reddit scraper—registering a developer account, installing packages via Homebrew, and running a Python script—that yielded 34,000 rows of AI agent conversations, then had Claude analyze it into a weighted table of top user themes.
- To prepare for product interviews, Klein uses ChatGPT-4 Voice Mode with a custom GPT that simulates unbiased product sense questions and delivers candid feedback, replicating realistic interview conditions.
Imagine with Claude | On The Edge #5
All About AI • October 06, 2025
This video showcases “Imagine with Claude,” an AI feature that builds complete software interfaces and applications on demand—including a file explorer, Minesweeper game, spreadsheet, and paint program—entirely in real time without any prewritten code.
Key Takeaways:
- Imagine with Claude dynamically builds entire software interfaces (OS-like UI, file explorer, text editor) in real time without any prewritten code or predetermined functionality.
- On demand, Claude can generate interactive applications such as a mind_sweeper.exe game—including a playable Minesweeper grid—though initial implementations may deviate from classic rules.
- It can spawn productivity tools like an Excel-like spreadsheet that executes formulas (e.g., summing cells A1 to A3) and graphic editors that draw complex images (a parrot and a horse at sunset) directly on the canvas.
.agent folder is making claude code 10x better...
AI Jason • October 06, 2025
AI Jason shows how to expand Claude Code’s 200,000-token context by structuring a .agent folder—housing PRDs, system docs, SOPs, and a readme—while pruning unused MCPS, offloading research to sub-agents, and automating updates with /update-doc.
Key Takeaways:
- Claude Code’s default context limit is 200,000 tokens, and removing unused MCPS from an agent instantly recovered about 2% of that window.
- Delegating isolated research steps to sub-agents and returning only summarized results minimizes noise and conserves main-thread tokens.
- A standardized .agent folder—organized into task (PRDs), system architecture, SOPs, and readme—plus an /update-doc command ensures up-to-date, concise codebase documentation for efficient context engineering.