Startup CEOs who’re “tokenmaxxing” are bragging that they’re spending more cash on AI compute than it could value to rent human employees. Astronomical AI payments are actually, in a sure nook of the tech world, a supposed marker of progress and success.
“Our AI invoice simply hit $113k in a single month (we’re a 4 particular person crew). I’ve by no means been extra pleased with an bill in my life,” Amos Bar-Joseph, the CEO of Swan AI, a coding agent startup, wrote in a viral LinkedIn put up just lately. Bar-Joseph goes on to clarify that his startup is spending cash on Claude utilization payments fairly than on salaries for human beings, and that the corporate is “scaling with intelligence, not headcount.”
“Our purpose is $10M ARR [annual recurring revenue] with a sub-10 particular person org. We don’t have SDRs [sales development representatives], and our paid advertising and marketing funds is zero,” he wrote. “However we do spend a sh*t ton on tokens. That $113K invoice? Part of it IS our go-to-market crew. our engineering, assist, authorized.. you get the purpose.”
A lot has been written in the previous few weeks about “tokenmaxxing,” an arrogance metric at tech startups and tech giants by which the sum of money being spent on AI instruments like Claude and ChatGPT is seen as a measure of productiveness. The Data reported earlier this month on an inside Meta dashboard referred to as “Claudenomics,” a leaderboard that tracks the variety of AI tokens particular person staff use. The overall narrative has been that the extra AI tokens an worker makes use of, the extra productive they’re and the extra progressive they have to be in utilizing AI.
Tales abound of particular person staff spending a whole lot of 1000’s of {dollars} in AI compute by themselves, and this being one thing that different employees ought to aspire to. There was no less than a partial backlash to this, with Salesforce saying they’ve invented a metric referred to as “Agentic Work Items” that makes an attempt to quantify whether or not all this spend on AI tokens is translating into precise work.
Shifting a lot cash and a focus to utilizing AI instruments is, in fact, being executed with the purpose of changing human employees. We have now seen CEOs justify mass layoffs with the concept that bettering AI effectivity will scale back the necessity for human employees, and Monday Verizon CEO Dan Schulman mentioned he expects AI to result in mass unemployment.
However whereas massive firms are utilizing AI to justify lowering employee headcount, startups are utilizing AI to justify by no means hiring human employees within the first place.
“That is the half individuals miss about AI-native firms – the $113k just isn’t a price, it’s your headcount funds allotted in another way,” Chen Avnery, a cofounder of Fundable AI, commented on Bar-Joseph’s LinkedIn put up. “We run an analogous mannequin processing mortgage paperwork that will usually require a crew of 15. The mathematics works when your AI spend generates 10x the output of equal human value. The actual unlock is compound scaling—token spend grows linearly whereas output grows exponentially.”
Medvi, a GLP-1 telehealth startup that has two staff and 7 contractors was constructed largely utilizing AI, is outwardly on monitor to herald $1.8 billion in income this 12 months, in line with the New York Occasions (Medvi is dealing with regulatory scrutiny for its practices). The trade has turn out to be obsessive about the thought of a “one-person, billion-dollar firm,” and numerous AI startups and enterprise capital corporations are actually making an attempt to push founders to attempt to create “autonomous” firms which have few or no staff.
Andrew Pignanelli, the founding father of the dubiously-named Basic Intelligence Firm, gave a presentation final month by which he defined that lots of the “jobs” at his firm are only a collection of AI brokers, and that he now often spends more cash on AI compute than he does on human salaries.
“We’ve began spending extra on tokens than on salaries relying on the day,” he mentioned. “As we speak we spent $4 grand on [Claude] Opus tokens. Some days it’ll be much less. However this reveals that we’re beginning to shift our human capital to intelligence.”
What’s left unsaid by these tokenmaxxing entrepreneurs, nonetheless, is whether or not the spend on AI compute is definitely value it, whether or not the cash could be higher spent on human staff, what sorts of disasters may happen, and whether or not any of that is truly financially sustainable.
Firms like OpenAI and Anthropic are shedding tons of money on their merchandise; although synthetic intelligence compute is dear, it’s underpriced for what it truly prices, and it’s not clear how lengthy traders in frontier AI firms are going to be keen to subsidize these losses. In the meantime, we now have reported endlessly on “workslop” and the human cleanup that’s typically wanted when AI-written code, AI-generated work, and customer-facing AI merchandise go awry. There are additionally quite a few horror tales of AI getting caught in a loop and burning 1000’s of {dollars} value of tokens on what find yourself being fully ineffective duties. Regardless, there’s a wholly new class of entrepreneur who appears hell-bent on “hiring” AI staff, not human ones.
Concerning the creator
Jason is a cofounder of 404 Media. He was beforehand the editor-in-chief of Motherboard. He loves the Freedom of Data Act and browsing.
