Corey Knowles

Yeah, I think that's evolving.

I think being more of a generalist maybe, but even if you want to be an expert in a certain field, I think you can also be very accelerated.

We use Codex at OpenAI.

I'm a generalist in my role.

I'm doing the easier side of code changes, really easy ones for what's getting deployed, medium ones for what's getting prototyped, just trying stuff.

I use it a ton.

to answer questions about the code base so i don't have to bother engineers it's like me and folks like me on like product teams are using ai and codecs a ton for those kinds of use cases um yeah they're engineers right and they're like more specialized of a function and they're using codecs to just be like massively accelerated like they gave the example of like sora like atlas was like some of the top codecs users internally at openai i mean like basically you know

nearly all of technical staff at OpenAI is using Codex.

And engineering, it's like a massive accelerant for us.

And there was a time where Codex usage was like growing, but it was like, went from like a little bit under 50% to like over 90%.

I forget the exact months, but it was like earlier this year, like spring or summer.

And then like,

Yeah.

And then all the Codex models after that were really big for us.

And we improved the product a ton as well.

And so there was a time where we were just noticing like, wow, the people working with Codex are making like 70% more PRs per time period.

And that's not like that's the best way to measure engineering productivity, but it was an easy thing that we could just pull.

And so we saw productivity go up a ton there within engineering and like then projects like Atlas, which we just launched have been like, there's like amazing anecdotes of like engineers saying like, yeah, what used to take like maybe like three engineers, three weeks is now one engineer, one week.

Wow.

So it's, we've been working so much faster thanks to that.