Most companies are treating their AI token bill as a technology cost. It isn't. It's a payroll decision wearing a developer invoice's clothes.

That distinction matters more than almost anything else in how you should be thinking about AI right now.

Here's where the confusion starts. When Eric Siu and Neil Patel sat down to discuss OpenClaw founder Peter Steinberger's reported $1.3 million annual token spend, the number landed in the marketing world like a data point and an anomaly in equal measure. One person spending that much on AI tokens? But dig into what that $1.3 million actually buys, and the framing shifts entirely. Steinberger isn't buying software licenses or headcount. He's buying execution capacity that scales without proportional hiring. That's payroll replacement, not infrastructure spending.

The market is starting to catch up to this distinction, and not gracefully. New data from Ability.ai shows organizations burning through thirteen times more AI tokens than last year, with the majority of that spend running through systems with zero measurable ROI tracking. The Pragmatic Engineer confirmed the pattern in a late April piece: most engineering teams consuming AI tokens at scale have no coherent framework for whether the spend is generating value proportional to its cost. They're not measuring. They're just spending.

This is the part that should make every founder and CMO flinch. Token prices aren't staying cheap.

The math is about to get uncomfortable.

Average cost per million tokens across major providers fell from roughly $10 to $2.50 in a single year according to Ramp's enterprise data cited by Investing.com. That collapse feels like good news for AI adoption. It isn't, not at scale. What it actually signals is that the subsidized window, underwritten by VC capital and provider growth incentives, is closing. The infrastructure economics that drove $2.50/million tokens won't persist as demand outpaces subsidized capacity. The organizations that are building their operational rhythm on untracked token spend are going to get a very unpleasant budget surprise in the next twelve to eighteen months.

Meanwhile, the real question nobody is asking in the rush to deploy AI everywhere is: what are you actually buying?

Taste doesn't come in an API call.

The Marketing School episode landed on something that gets dismissed as soft because it's genuinely difficult to quantify. Experience and taste outperform raw AI output in marketing and creative work. Not sometimes. Consistently. That $1.3 million person isn't winning because he has access to more tokens than his competitors. He's winning because his team has the judgment to know which outputs are worth shipping, which angles land, which positioning feels right versus which sounds generated.

Brian Chesky keeps coming up in this conversation for good reason. Player-coach leaders, the ones who stay in the arena doing the work rather than managing managers, are the ones shipping anything worth looking at. Small teams with strong taste and AI leverage are delivering disproportionate results because the leverage amplifies good judgment. AI amplifies what you bring to it. It cannot substitute for the thing you're bringing.

This is where most token-spend conversations go wrong. Companies are buying model access and expecting good output without understanding that the output is only as good as the taste encoded in the instructions, the approval loop catching the bad calls, and the orchestration layer holding the whole system together coherently.

That orchestration layer, the thing that gives you logs, approvals, rollback, and visible state — that's what turns expensive token consumption into a productive workflow. Without it, you're just burning money on probabilistic autocomplete.

The actual decision is this

You are not buying AI tokens. You are buying a workforce that runs on your behalf. Your token budget is a headcount decision. Every call you make against that budget either generates value or it doesn't. The companies that are going to win this next phase are the ones treating AI spend like payroll — with the same rigor, measurement, accountability, and strategic clarity they apply to every other workforce decision.

If you can't explain what your AI token spend is producing, you don't have an AI strategy. You have a bill.

The reckoning is coming. When it arrives, the companies with taste, judgment, and orchestration discipline are going to look very smart. The ones who just bought more tokens won't.

---

Foundry Works helps businesses build AI operations that actually produce results. If your token spend needs a strategy behind it, you know where to find us.