Carl Franzen

BREAKING: Disney is reportedly exiting the deal it signed with OpenAI, in which it pledged to invest $1 billion in the company and agreed to license some of its characters for use in Sora, per the Hollywood Reporter

Sora is being discontinued. There were rumors, but I assumed this was going to be limited to the app. Looks like this is comprehensive however, even the API. This is sad to see for me personally.

"Your job won't be taken by AI, it will be taken by Meek Mill using AI" - Jensen Huang

HOW DID I REACH MY LIMITS ON CLAUDE THIS MORNING AFTER SENDING 3 MESSAGES WTF IM ON A MAX PLAN

We’re saying goodbye to the Sora app. To everyone who created with Sora, shared it, and built community around it: thank you. What you made with Sora mattered, and we know this news is disappointing. We’ll share more soon, including timelines for the app and API and details on preserving your work. – The Sora Team

Three months ago, my daily AI use was like 80% ChatGPT / 15% Claude / 5% Gemini. Now it's like 80% Claude / 15% ChatGPT / 5% Gemini. What happened? I think it was gradual, then sudden: 1. I tested models head-to-head for a month. Claude and Gemini kept beating ChatGPT on my typical use cases. But the Gemini app has no agentic harness (can't work with files in there) and I don't like its privacy options. So I started shifting to Claude. 2. I have a strong preference for Claude's writing style and artifacts. I just find both more aligned to my personal preferences. I like reading narrative, not scanning bullets. I couldn't get ChatGPT to write in a way that I enjoyed. Even with GPT-5.4, I still can't get this consistently. I can torture it into using fewer bullets, but then it doesn't use bullets when it should. And it still feels so unnatural, like always ending with an offer to do something. Just feels like a template, a machine, not personable, not like a colleague I want to work with. And I like how Claude displays artifacts in a big window side-by-side with my chat. I find this really useful for visual reviews. So I started using Claude more. 3. I then started having more need for good design from my AI chatbots. This includes PowerPoint presentations, web UIs, and even print designs once I discovered that I could get Claude to create those with WeasyPrint and Prince XML. I was hopeful that GPT-5.4 would be much better at PowerPoint in particular than prior GPT models, since we create so many presentations at work. Unfortunately, this wasn't the case. I've tried everything OpenAI suggested in its recent blog post, but still can't get consistently good design, whereas Claude Opus 4.6 just has an innately good design sensibility. 4. Right as my preference started to shift to Claude, Anthropic started to build out Cowork. I had previously used Claude Code, but Cowork made it so easy to just work in any local folder without worrying about things like dependencies. This was really a big turning point. Since I already preferred Claude's writing, design, and document creation, Cowork let me put all those pieces together to expedite work and expand my ambitions for leveraging AI. 5. Just as I was finding this, Claude took off at work. Many colleagues started having similar feelings to me. Demand for Claude licenses started to increase, even though we all have access to Gemini in Workspace and ChatGPT Enterprise. Our policy is to give broad access to the best available AI tools, so as people demonstrated use cases that Gemini and ChatGPT didn't address as well, more people started using Claude. This creates a feedback loop: More Claude use leads to more identification of areas Claude excels, which leads to more Claude use. 6. I was very hopeful when Codex started getting good. I still think Codex is excellent for when I'm building complex applications with no frontend. But because I don't want to jump back and forth between different tools, and because Claude is so good at frontend, I just find myself mostly sticking to Claude. I do find Codex marginally better at some things I've tried, but the switching cost of going to Claude for frontend and Codex for everything else just hasn't been worth it. 7. I increasingly find OpenAI's priorities misaligned with mine. I want a tool that makes me more more productive and is enjoyable to interact with. But I find most announcements from OpenAI (with the notable exception of the Codex and Atlas teams) aren't that. And features that are important to work use cases, like document syncing in projects, for example, still haven't made it to our Enterprise account. Meanwhile, Anthropic seems to have a more intuitive sense of what will make me more productive. For example, Cowork projects, scheduled cowork tasks, scheduled cloud code tasks. They keep rolling stuff out that I'm immediately finding useful and using. 8. Most people I've met at OpenAI really care about building great models and products, and genuinely want constructive feedback, including criticism. I'm pretty sure I've directly shared most if not everything above with multiple people at the company and have received interest and respectful replies, and often on X engineers will read a criticism and take action. But I also sometimes get the sense that having almost a billion users is justification for continuing with the status quo. For example, when addressing Anthropic's Super Bowl ad, OpenAI reps kept repeating the line about having different challenges from Anthropic because they had so many more users. That's true, but to me it also felt kind of alienating and dismissive; revenue is probably a more important indicator of value than usage, and Anthropic's revenue per user is way higher (yes, they have way fewer free users, which helps explain that, but it's also kind of the point, because they're building something people want to pay for). None of this is to say my usage won't shift again. I find Atlas to be useful (though use it less now that I'm in Claude more, since Claude uses a Chrome extension), I find Codex to be solid, I find GPT models excellent at search and hard reasoning tasks. But when I saw Fidji push for OpenAI to focus on enterprise, my thought was: good idea, and thank goodness. Now let's see where it leads.