High Signal Podcasts Evidence ledger
Method
Browse
← Back to evidence

Evidence receipt / commitment

Published · transcript-backed

Alexander Embiricos: commitment

14 Dec 2025 Lenny's Podcast Why humans are AI’s biggest bottleneck (and what’s coming in 2026) | Alexander Embiricos (OpenAI Codex Product Lead)

“I think we will very much get to the point, well, basically where the boundaries are a little bit blurred.”

— Alexander Embiricos

Source trail

Everything needed to verify it.

Speaker
Alexander Embiricos
Attribution
Verified speaker
Claim type
commitment
Recorded
14 Dec 2025
Publisher
Lenny's Podcast

Transcript context

…Do you think this eventually moves to non-engineers doing this sort of thing? Does it have to be an engineer building this thing? Could Sora have been built by, I don't know, a PM or designer? I think we will very much get to the point, well, basically where the boundaries are a little bit blurred. I think you're going to want someone who understands the details of what they're building, but what details those are will evolve. Kind of like how now if you're writing Swift, you don't have to speak assembly. There's a handful of people in the world, and it's really important that they exist and speak assembly, maybe more than a handful, but that's a specialized function that most companies don't need to have. So I think we're just going to naturally see an increase in layers of abstraction. And then the cool thing is now we're entering the language layer of abstraction, like natural language, and then natural language itself is really flexible. You could have engineers talking about a plan and then you could have engineers talking about a spec, and then you could have engineers talking about just a product or an idea. So I think we can also start moving up those layers of abstraction as well. But I do think this is going to be gradual. I don't think it's going to go off to all of a sudden nobody ever writes anything, any code and it's just specs. I think it's going to be much more like, "Okay, we've set up our coding agent to be really good at previewing the build or at running tests," maybe that's the first part that most people have set up. And it's like, "Okay, now we've set it up so they can execute the build and it can see the results of its own changes, but we haven't yet built a good integration harness so that it can," in the case of Atlas... By the way, I don't know if they've done any of this or not. I think they've done a lot of this. But maybe the next stage is enable it to load a few sample pages to see how well those work. So then, okay, now we're going to set it up to do that. And I think for some time at least, we're going to have humans curating which of these connectors or systems or components that the agent needs to be good at talking to. And then in the future, there will be an even greater unlock where Codex tells you how to set it up or maybe sets itself up in a repo. What a wild time to be alive. Wow. I'm curious just the second order effects of this sort of thing, just how quickly it is to build stuff. What does that do? Does that mean distribution becomes much, much more important? Does it mean ideas are just worth a lot more? It's interesting to think about how quick how that changes.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.

Search evidence