OpenAI ran DevDay this week, and within hours my feeds had sorted themselves into the two usual camps: this changes everything, and this changes nothing. As ever, the truth was sitting quietly in the middle being ignored by both.
I watched it because it's my job to know roughly what's coming, not because I expected to be moved. The agent-building tooling and the app stuff are the headline, and the headline is the part I trust least. Demos are choreography. They show you the happy path with the network behaving and the prompt rehearsed. None of that tells you how it behaves at 4pm on a Friday when the rate limits bite and the model decides to be confidently wrong about something load-bearing.
What I actually care about is the boring layer underneath: the API surface, the pricing, the latency, whether the thing is stable enough to put in front of real users without an apology in the release notes. That's the bit you only learn weeks later, after the keynote glow has worn off and people are filing the bug reports.
So I'm not going to have an opinion yet, which apparently makes me a coward. I'll wait until I've actually built something with it and watched it fall over. That's when you find out what a keynote was really worth.