Evidence receipt / evaluation
Published · transcript-backedAlexander Embiricos: evaluation
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 like to joke today that AI products, and it's a half joke, they're actually really hard to use because you have to be very thoughtful about when it could help you.”
Source trail
Everything needed to verify it.
- Speaker
- Alexander Embiricos
- Attribution
- Verified speaker
- Claim type
- evaluation
- Recorded
- 14 Dec 2025
- Publisher
- Lenny's Podcast
Transcript context
…I thought the point about not reading Slack and Datadog was it's just not distracted, it's just constantly focused and is always in flow. But I get what you're saying there is it doesn't have all the context on everything that's going on. Yeah. And that's not only true when it's performing a task, but again, if you think of the best team and teammates, you don't tell them what to do. Maybe when you first hire them, you have a couple meetings and you're like, "Hey," you learn, "Okay, these prompts work for this teammate, these prompts don't. This is how to communicate with this person." Then eventually you give them some starter tasks, you delegate a few tasks. But then eventually you just say like, "Hey, great. Okay, you're working with this set of people in this area of the code base. Feel free to work with other people on other parts of the code base too, even. And yeah, you tell me what you think makes sense to be done." And so we think of this as proactivity and one of our major goals with Codex is to get to proactivity. I think this is critically important to achieve the mission of OpenAI, which is to deliver the benefits of AGI to all humanity. I like to joke today that AI products, and it's a half joke, they're actually really hard to use because you have to be very thoughtful about when it could help you. And if you're not prompting a model to help you, it's probably not helping you at that time. And if you think of how many times the average user is prompting AI today, it's probably tens of times. But if you think of how many times people could actually get benefit from a really intelligent entity, it's thousands of times per day. And so a large part of our goal with Codex is to figure out what is the shape of an actual teammate agent that is helpful by default. When people think about Cursor and even Cloud Code, it's like a IDE that helps you code and auto completes code and maybe does some agentic work. What I'm hearing here is the vision is different, which is it's a teammate. It's like a remote teammate, a building code for you that you talk to and ask to do things. And that also does IDE, auto complete and things like that. Is that a kind of a differentiator in the way you think about Codex?…
Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.