Uber and Meta have said no to toxenmaxxing. Make way for 'yieldmaxxing'

And the reason we are talking about limits, do you know why we are talking about limits, Vidhatri? Because two things have happened over the last couple of weeks.

I want you to tell me what is it that happened that, you know, g- got to this episode.

Why are we doing this episode?

Okay.

So I think the first thing is Meta's recent announcement, the back and forth that happened.

I remember, uh, sometime last year, I remember reading that they had a leaderboard of sorts where they were tracking their employees' AI usage, and I think New York Times, uh, had this article where it said that AI, um, uh, uh, employees are maxing out, and everybody then latched onto that word and token maxing became a thing.

Um, and everybody was just talking about how, how many tokens they were using.

But with everything in AI, the conversation seems to be moving really fast.

Now, there's a flip to this whole situation.

Meta has said, um, "Let's stop this leaderboard, and let's actually review our AI usage properly and not just reward people for using the maximum number of tokens."

That's correct.

So they started off with a leaderboard where they said, "Look, maximum AI users." So I think people started using AI and the latest and the greatest models to do everything from probably, like, ordering a coffee to, like, pushing out really complex code to doing complex, like, product and orchestration.

Let's create an agent for this, let's create an agent for that, and things went out of hand.

And this is not just Meta, by the way.

There are two other companies, all of whom suddenly, suspiciously, curiously, in the same period of time realized, "Hey, this AI thing is a little expensive, guys.

We need to do something about it." Uh, one of them was, of course, Meta.

The second was Microsoft.

So Microsoft actually had GitHub Copilot, which is their in-house, you know, AI tool.

They're one of the pioneers of it.

Then they started using Claude Code, and just a few weeks back, there were, like, a couple of divisions inside Microsoft where they said, "Mm, I don't think we should use Claude anymore.

Please let's go back to GitHub Copilot." So they started doing that.

And the third company, and in fact, this was the company that actually, to their credit, started off this entire conversation and kicked the ball down the road, was Uber.

And there were not one, not two, but three separate executives over a period of time who came and said, "You know what? We are using a lot of AI." In fact, one of them said that, "We blew through our entire annual AI budget in just, like, four months, and now we have questions about whether this whole thing is worth it or not, because we don't know if the outcomes are there or the outcomes are not there." That's the first big theme that's going on, where companies, especially big tech companies, are now looking at their AI budgets and saying, "Damn, this is getting out of hand." But there is a second thing that happened, and that second thing that happened was happening just over the last few days, and that also has to do with Claude.