Are AI tokens the new signing bonus or just a cost of doing business?

This week, a subject that has been boomeranging round Silicon Valley bounced into the highlight: AI tokens as compensation. The thought is simple sufficient — reasonably than giving engineers solely wage, fairness, and bonuses, corporations would additionally hand them a finances of AI tokens, the computational items that energy instruments like Claude, ChatGPT, and Gemini. Spend them to run brokers, automate duties, crank by means of code. The pitch is that entry to extra compute makes engineers extra productive, and that extra productive engineers are price extra. It’s an funding within the individual holding them, is the thought.

Jensen Huang, the leather-jacket-wearing CEO of Nvidia, appeared to seize everybody’s creativeness when he floated the notion on the firm’s annual GTC occasion earlier this week that engineers ought to obtain roughly half their base wage once more — in tokens. His prime individuals, by his math, would possibly burn by means of $250,000 a yr in AI compute. He known as it a recruiting instrument and predicted it could turn into commonplace throughout Silicon Valley.

It isn’t fully clear the place the thought was first, nicely, ideated. Tomasz Tunguz, a famend VC within the Bay Space who runs Idea Ventures and focuses on AI, information, and SaaS startups — and whose writing on all issues information has garnered a loyal following through the years — was speaking about this in mid-February, writing that tech startups had been already including inference prices as a “fourth element to engineering compensation.” Utilizing information from the compensation monitoring website Ranges.fyi, he put a top-quartile software program engineer wage at $375,000. Add $100,000 in tokens and also you’re at $475,000 totally loaded — that means roughly one greenback in 5 is now compute.

That’s no coincidence. Agentic AI has been taking off, and the discharge of OpenClaw in late January accelerated the dialog significantly. OpenClaw is an open-source AI assistant designed to run repeatedly — churning by means of duties, spawning sub-agents, and dealing by means of a to-do record whereas its person sleeps. It’s a part of a broader shift towards “agentic” AI, that means techniques that don’t simply reply to prompts however take sequences of actions autonomously over time.

The sensible consequence is that token consumption has exploded. The place somebody writing an essay would possibly use 10,000 tokens in a day, an engineer working a swarm of brokers can blow by means of tens of millions in a day — routinely, within the background, with out typing a phrase.

By this weekend, the New York Occasions had put collectively a wise have a look at the so-called tokenmaxxing pattern, discovering that engineers at corporations together with Meta and OpenAI are competing on inside leaderboards that observe token consumption. Beneficiant token budgets are quietly turning into an ordinary job perk, the paper reported, the best way dental insurance coverage or free lunch as soon as was. One Ericsson engineer in Stockholm instructed the Occasions he most likely spends extra on Claude than he earns in wage, although his employer picks up the tab.

Perhaps tokens actually will turn into the fourth pillar of engineering compensation. However engineers would possibly need to maintain the road earlier than embracing this as a simple win. Extra tokens could imply extra energy within the brief time period, however given how briskly issues are evolving, it doesn’t essentially imply extra job safety. For one factor, a big token allotment comes with giant expectations. If an organization is successfully funding a second engineer’s price of compute in your behalf, the implicit stress is to supply at twice the speed (or extra).

And there’s a muddier drawback beneath that: on the level the place an organization’s token spend per worker approaches or exceeds that worker’s wage, the monetary logic of headcount begins to look completely different to its finance workforce. If the compute is doing the work, the query of what number of people have to be coordinating it turns into tougher to keep away from.

Jamaal Glenn, an East Coast-based Stanford MBA and former VC turned monetary companies CFO, equally factors out that what could seem to be a perk generally is a intelligent method for corporations to inflate the obvious worth of a compensation package deal with out rising money or fairness — the issues that truly compound for an worker over time. Your token finances doesn’t vest. It doesn’t admire. It doesn’t present up in your subsequent supply negotiation the best way a base wage or fairness grant does. If corporations efficiently normalize tokens as pay, they could discover it simpler to maintain money comp flat whereas pointing to a rising compute allowance as proof of funding of their individuals.

That’s deal for the corporate. Whether or not it’s deal for the engineer will depend on questions most engineers don’t but have sufficient data to reply.