Token Maxxer 1%
5 Idea Friday: ai powerusers; no bubbles here; K becomes E; mind the supercore; no country for youngs
5 Idea Friday
ai agent usage, and the emerging top 1%
bubbles? no bubbles…indecision, momentum, hunting for the triple-digit returns, high premium to growth, but bubbles?
the K becomes an E
mind the supercore (reprise, ad infinitum)
no country for youngs, or the education bag-holder
👉👉👉Reminder to sign up for the Weekly Recap only, if daily emails is too much. Find me on twitter, for more fun. 👋👋👋Random Walk has been piloting some other initiatives and now would like to hear from broader universe of you:
(1) 🛎️ Schedule a time to chat with me. I want to know what would be valuable to you.
(2) 💡 Find out more about Random Walk Idea Dinners. High-Signal Serendipity.1. The Token-Maxxer 1%
One of the more fun things about the “new media” turn, “go direct,” and the general disintermediation of tradmedia when it comes to company-storytelling is that a lot of high profile companies are hiring in-house economists, and those economists publish fun stuff with all the data that these companies have.
Anthropic, for example, just released a report on all kinds of things Claude-related, chock full of fun nuggets like this one:
It’s kind of amusing that “casual conversation” soon gives way to “sleep advice,” which peaks at ~7am, at which point it’s plainly too late to seek advice.
This, too, was amusing, in the sense of extremely on-the-nose stereotype reinforcement:
When it comes to using Claude, men tend to get work done, code, and build automations, while women . . . prefer to chat. But, it’s a signal of “collaborative engagement,” says Anthropic. Lol.
To be fair, women were only 12% of the sample, but still, it’s funny.
OAI, for its part, also released a report, focused on the “shift to agentic AI.” It’s also chock full of interesting stuff.
The gist of the thing is that more people across more functions are using Codex (rather than the web interface), and that token outputs are rising alongside (implying more agent usage). Naturally, OAI employees have way-above normal usage trends, but outside the org, similar patterns are visible.
The most interesting observation, however, is this one:
Putting aside the most recent June data, while everyone is running more agents (measured by cumulative daily active runtime), the top centile of users are really running more agents.1
There’s a step-change at the tail end of the distribution, where the gap is both substantial, and growing.
Here’s another cut of a similar observation:
Per Ramp’s spending data (so take it for what it is), the top 1% of token spend/employee is more than 10x the top 10%. Once again, the gap is big and growing.
All of which is to say there’s more fuel to the “AI Power User” fire.
The tech is real and transformative, but a small number of users appear to be way out in front, and they are compounding their lead. And that’s the thing about compounding exponentials: small differences become very big differences very quickly.
And therein lies some of the trickiness of finding “AI winners,” or the broader question of observing “AI ROI” in the 10,000ft picture. (1) If gains are concentrated then, by definition, diffuse benefits will be hard to observe (at first), widespread margin growth, notwithstanding; and (2) if you thought we had pareto distributions before, well, you ain’t seen nothing yet.
ICYMI
2. Not Bubbly. Indecisive, volatile, choosy, but not bubbly
Returning briefly for a moment to the totally-not-bubbly-but-rather-highly-discerning-and-somewhat-uncertain stock market, considering the following.
The Mag7 continued their underperformance relative to the Nasdaq100, but this week, they took it to a whole new level:
