I saw some clips from the OpenAI Dev Day 2025. What follows is a little stream of consciousness as I read through OpenAI’s announcement blog posts.
Apps in ChatGPT – OpenAI is looking for ways to capture more eyes for longer. This is a move to further replace Google as a home screen. Google wishes they’d done this first. Ads will follow and the naked pay-to-play model will expand. To be a little snarky, it’s notable that the first apps on display are.. not interesting. It’s iframed content from another site. It’s nice that ChatGPT is sending traffic to the other sites, but the convenience or time savings here are essentially zero. Curious to see what happens to this in the future.
Introducing AgentKit – The capabilities look interesting. Adding visual agent workflows really does make the process more accessible for small and medium companies, and for a wider variety of stakeholders. Fitting the flows onto PPT slide is a basic feature expected by SaaS customers. But pulling all the key features (Evals and RFT in particular) into “one place” was a really smart move. People were already moving in to provide these kinds of turn-key ETE services. OpenAI saw what they were doing and just.. built it in house.
Sora in the API – It’s hard to hide my personal bias here… generative AI video scares me. Making it widely available for embedded and automated applications via the API scares me more. The opportunity to generate high volumes of AI video is much larger now.
Codex now generally available – I am not a dev, but I know a lot of devs. I heard a decent amount about Codex before this announcement. Most notable is the direct Slack integration. The idea that developers can start a new conversation about something, then tag in Codex a little later is kind of huge. It’s a new way of thinking about integrating the intelligence not just into workflows, but into the business. We’ve all known that this is where agentic processes and chat bots are headed.. and now we’re starting to see just how OpenAI can remove the need for new windows and entry methods.
…
The three other announcements (GPT-5 Pro in the API, gpt-realtime-mini, and gtp-image-1-mini) are rinse-repeat.