Okay, so this is genuinely one of the funniest things I’ve read all year, and I have to talk about it because we have officially entered Clown World™ when it comes to enterprise AI.
Companies — like, real companies, with real CFOs and real spreadsheets, are now spending more money on AI agents than they would just pay their human employees. Not “comparable to.” Not “approaching.” More. That’s where we are. That’s the timeline we’re in.
Let me walk you through this, because it’s wild.
The Setup: AI Agents Are Spicy Little Tokens-Burning Machines
Here’s the deal. AI agents are great at cranking out massive amounts of work, especially code. Like absurd amounts. Engineers will literally fire up multiple agents at once, set them loose on different tasks, walk away, get a coffee, come back, and there’s just a mountain of code waiting for them.
Sounds amazing, right? Productivity through the roof. Cyberpunk dream. The future is here.
Yeah. About that.
Every single one of those requests costs tokens. And tokens cost money. And when you’ve got an entire engineering team running agents 24/7 like it’s a Bitcoin mining rig that occasionally writes Python, the bill gets… let’s call it “spicy.”
Bryan Catanzaro, who’s the VP of applied deep learning at Nvidia, a company that, last I checked, is not exactly hurting for cash, straight up told Axios: “For my team, the cost of compute is far beyond the costs of the employees.”
Read that again. The compute costs more than the humans. At Nvidia. The company selling the compute.
The Companies Drinking Their Own Kool-Aid
Here’s where it gets even more chef’s kiss. The companies building these AI tools are also the ones using them the most aggressively, which means they’re getting hit hardest by their own product.
Anthropic, the Claude people, said earlier this year that “pretty much 100 percent” of their code is now AI-generated. One hundred percent. Google and Microsoft execs say it’s about a quarter at their shops. And over at Meta, employees are now getting performance-reviewed based on how much AI they use. So if you’re not burning tokens, you’re underperforming. Got it. Cool. Totally normal.
This is basically corporate peer pressure with a price tag attached.
Tokenmaxxing Is a Real Thing and I Hate It
Okay, this part. THIS PART.
Apparently — and I’m sorry to be the one to tell you this — the slang for power-users who burn through tokens is “tokenmaxxing.” Yes, like that. Yes, with all the connotations you’re thinking of. No, I don’t make the rules.
Some of these guys are racking up monthly bills north of $150,000. A software engineer in Stockholm told The New York Times, and I quote, “I probably spend more than my salary on Claude.” Imagine looking your boss in the eye and saying that with a straight face.
And get this — Uber engineers using Claude Code already blew through the company’s entire 2026 AI budget. We’re in April. APRIL. They torched the whole annual budget in four months. That’s not a budget, that’s a starter pack.
Jensen Huang’s Galaxy-Brained Solution
So how are tech leaders responding to this? With completely normal, measured business decisions, obviously.
Just kidding. Nvidia CEO Jensen Huang’s pitch is to give engineers AI tokens worth roughly half their base salary — and use it as a recruiting tool. Like, “forget signing bonuses, come work for us and you’ll get to use SO MUCH AI.”
I’m sorry, what? “Hey new hire, instead of money, here’s a stipend to feed back to a different megacorp’s GPU farm.” That’s the pitch. That’s where we are.
The Cherry on Top
OpenAI investors are reportedly hyped about all this because they think Codex uses tokens more efficiently than Claude Code, so this is great news for their bag. Meanwhile, Anthropic just… raised prices. Because of course they did. When demand is “engineers in a feeding frenzy,” you don’t leave money on the table.
So everyone’s got their hand in this thing. The infrastructure providers are winning, the AI companies are winning, and somewhere in a CFO’s office, someone is staring at a quarterly compute bill that costs more than their entire engineering team’s salaries combined and slowly closing their laptop.
Does This Even Work Though?
Here’s the kicker — the part that should be the headline. There’s still no clear evidence this is actually better than just having humans do the work. AI agents are error-prone. They’ve already caused incidents at Meta and Amazon. Multiple studies suggest forcing employees to use AI tools is actually making their jobs harder, not easier.
So to recap: companies are spending more on AI than they would on humans, the AI is less reliable than humans, the humans hate using it, and somewhere a Stockholm engineer is spending his entire paycheck feeding the machine.
I genuinely cannot wait to see how this looks in five years. Either we’ll all be tokenmaxxing on our personal AI swarms, or someone’s going to look at the books and have a very, very bad day.
Probably both, honestly.
