August 2025: A few AI things that inspired me this month

Recently, I have been writing about and discussing how we design the next wave of AI systems. And not just the models themselves, but the people, organizations and interfaces that make the systems useful. From first-hand accounts of scaling OpenAI, to frameworks for User Agent Interfaces, to application-layer plays like Gamma, I think there is a lot to learn about the patterns we see emerging. There is no single handbook to follow for the AI era!

As always, this is a collection of a few articles, videos or podcasts that sparked great conversations with my team and helped me see things from a fresh perspective. These aren’t necessarily product announcements or major headlines, just ideas that resonate. I hope you find them useful too.

📖 Read: “Reflections on OpenAI”

This is a fascinating first-hand account from Calvin French-Owen (formerly at OpenAI) about what it is like to work inside one of the fastest-scaling AI companies. He covers everything from bottoms-up culture and Slack-driven collaboration to GPU economics and the intensity of rapid launches like Codex.

Really interesting to learn about the mix of ambition and improvisation – including how OpenAI operates with a bias for action, letting researchers act like a “mini-executive,” while still shifting direction when necessary. One clear takeaway is that scaling at this pace requires adaptability and a different playbook from a typical enterprise org.

 Read it here: Reflections on OpenAI | Calvin French-Owen

 🎧 Listen: “The Head of ChatGPT on AI Attachment, Ads, and What’s Next”

Nick Turley, Head of ChatGPT, talks about what it is like to run the fastest-growing software product ever. A key theme is that OpenAI thrives on ambiguity. Success inside the org depends on First Principles thinking and creative adaptability. Great complement to the previous article. The culture shift from execution to invention is one of the hardest transitions for leaders today.

Listen to the podcast here: The Head of ChatGPT on AI Attachment, Ads, and What’s Next | Decoder podcast

📖 Read: “UI vs API vs UAI”

This post outlines an interesting framework for what is currently happening with agents. First, we built a user interface (UI) for people. Then we made application programmable interfaces (APIs) so applications could be operated by other applications. Now, as reasoning agents begin operating software on our behalf, we need to design a new layer: User Agent Interfaces (UAIs).

The key is developing good design patterns (for example, feedforward, tolerance, feedback) that make systems legible to agents while still being operable by humans. It is early, but we can already see the outlines of new best practices. It shows UAIs as a parallel design surface we need to build with the same rigor as UIs or APIs.

Read it here: UI vs API vs UAI | Josh Beckman

🎥 Watch: “Scaling the 'Cursor for Slides' to $50M ARR”

This conversation with Gamma founder Jon Noronha is packed with insights. Gamma treats “design taste” as a core differentiator and invests heavily in designers because of how critical this work is for prompt building. They constantly A/B test across models to find the best fit for each task and share a unique view on their storytelling vision to create something beyond slide generation.

Watch it here: Scaling the 'Cursor for Slides' to $50M ARR ft Gamma founder Jon Noronha | Training Data podcast

 📖 Read: “What makes Claude Code so damn good (and how to recreate that magic in your agent)!?”

The TL;DR on this one is: “write less code and let the model cook.” Claude Code shows how architectural simplicity can be a superpower: one main loop, simple search, and a to-do list.

The author writes, “This is the Camera vs. Lidar of the LLM era — and I’m only half joking.” Claude Code makes smart choices between small, medium, and high-level tools – elevating common actions into explicit tools while using deterministic ones to stay on track. Really interesting view!

Read it here: What makes Claude Code so damn good (and how to recreate that magic in your agent)!? | minusx.ai