OpenAI DevDay 2026: Interview with David Ondrej

San Francisco, September 29. The OpenAI DevDay 2026 keynote had just ended. More than 20 launches: always-on agents, Codex in the cloud, new models, a $500 plan.

David Ondrej still had his badge in hand. He agreed to a few questions.

You were in the room. What was the vibe?

So this was my very first OpenAI DevDay, and I have to say... the vibe was excited, futuristic, like a new era was approaching.

There were A LOT of new releases, perhaps too many, because most of them didn't feel like they are for developers at all.

For me, the most useful announcement was probably Spaces - a new way of interacting with agents and your team - something like Google Docs, but built for the agent era.

Codex now runs in the cloud, even with your laptop closed. That sounds a lot like Cloudroom. Is OpenAI your competitor now?

In a way yes, however I'm not feeling anything. I don't see OpenAI's new announcement as any threat to Cloudroom, because our main offer is that we are free, open-source and agent-agnostic. You can use any model, any harness, any inference provider with Cloudroom. Plus, our GUI and Core are fully open-source. OpenAI will never open-source their Cloud infrastructure, and they will definitely not let you use Anthropic models on their platform. If anything, this validates the idea that Cloud Agents are the future.

You switched from Codex to Claude Code. After today, who's ahead?

Good question. In the AI space, however, this changes week to week. Before the release of Opus 5.5, it really did seem like OpenAI took the lead, but after the release of Opus 5.5 and Sonnet 5.5, now Anthropic is clearly in the lead. And not because these models are much better than Astra, but because of how fast and efficient they are -- I mean, with Opus 5.5 running at all times, I never even hit my Claude Code $200 limit. However, with GPT-6 Astra, I had six different $200 plans of Codex, and I used to hit limits in like one day of usage. So if I had to say... Anthropic is in the lead now. And the DevDay does not change that.

OpenAI launched Sign in with ChatGPT, then halved the limits on the $200 plan. What do you make of that?

Sure, of course we will support "Sign in with ChatGPT" and let people use the tokens they already pay for. What this weird release / upgrade / downgrade tells me, is that OpenAI is struggling for compute. They are losing to Anthropic when it comes to the value of the $200 subscription. They literally just launched a $500 subscription, which is the single most expensive subscription in all of AI. However, I have no need to upgrade right now. Why? Because Opus 5.5 from Anthropic is not only better than OpenAI's models, it's also faster, AND you get way more usage of it in the $200 Claude subscription. The $500 plan from ChatGPT simply does not make sense -- and even though we will let the Cloudroom users bring it -- I don't think many of them will upgrade to it.

Last one. Where are cloud agents in 12 months, and what's your next move?

In 12 months, Cloud Agents will be far more popular than today. In fact, it won't even be close. Cursor's own data from this year proves this: in January they only had about 10% of their production PRs coming from Cloud Agents. In June it was nearly 60%. However, Cursor is closed source and they have insane levels of ecosystem lock-in, so there will be many power users who will choose to look for alternatives... this is where Cloudroom comes into play.

To answer your second question: after attending my very first OpenAI DevDay my moves are quite literally the same. Cloudroom's trajectory is not impacted by this event whatsoever. The people who say that OpenAI is killing startup just lack creativity. It's a skill issues. In fact, tomorrow is Cloudroom's public launch -- we are launching as a fully free offering, letting anyone use Cloud Agents 100% free and open-source. So any power-user should go to Cloudroom.dev and join the waitlist.