Token Poker and Slop Cannons: How Companies Will Ration AI
The Primeagen's five predictions for the AI rationing era — token stipends, Token Poker, and 'slop cannons' — are half satire and completely plausible.
A developer's AI dashboard showing a token budget burning down toward zero before month's end.
I'm not a software developer — but I follow The Primeagen because nobody translates the chaos of software engineering culture into something this watchable. He's genuinely funny, consistently honest, and has inspired a lot of people (including myself to some extent) to learn coding and software engineering.
His latest prediction is one he isn't happy about: the wild west of AI coding — where engineers fire off agents like free ammo and nobody checks the bill — is ending. Not because the tools got worse, but because someone in finance finally read the invoice.
In a recent video bluntly titled "Unfortunately, I Was Right", he laid out five ways corporate software teams will start rationing tokens. They're half satire. They're also completely plausible — because the moment any resource gets metered, the bureaucracy that grows around it is depressingly predictable.
First, the receipts
This isn't hypothetical. Uber burned through its entire 2026 AI budget in four months, with per-engineer costs running $500 to $2,000 a month as Claude Code adoption jumped from 32% to 84% of its org. Microsoft canceled most of its direct Claude Code licenses after compute costs outran the cost of the employees using it. This is the unit-economics collapse I wrote about arriving as a line item — and once something becomes a line item, it gets managed.
Prime's five predictions
He breaks them into five distinct scenarios — none of them invented from thin air. Each one borrows a mechanic already common in how companies manage money, headcount, or cloud spend, and drops it onto token usage. Reading through them as someone who has never shipped a pull request in their life, I found them uncomfortably easy to picture.
- Token stipends. Like a 401k match, every engineer gets a yearly token budget — and a year-end bonus for coming in under it. Given those per-seat numbers are already real, this is the least speculative of the five.
- Token Poker. Planning Poker, but you estimate the token cost of a feature before you build it. Budget-conscious estimation becomes a mandatory ceremony, and consultants get a brand-new workshop to sell.
- Team and org budgets. One company-wide pool splits into departmental allocations, spawning a middle-management class whose entire job is lobbying for more tokens — plus "pair prompting," two engineers huddled over one cost-efficient prompt.
- Open-source token donations. Companies route a slice of their budget to fund critical OSS projects' CI/CD and AI maintenance. Prime calls this the heartwarming one, and he's right — the plumbing already exists. The Linux Foundation just announced $12.5M in grants from Anthropic, AWS, GitHub, Google, Microsoft, and OpenAI, and a new Tokenomics Foundation to standardize token-cost management.
- Rewarding lines of code with bigger budgets. The dystopian one: companies reward their most "prolific" AI coders — Prime's "slop cannons" — with larger token allocations, treating raw output as value.
The one that should actually worry you
That last prediction is the dangerous one, because it revives the worst metric in software history. Bill Gates said it best decades ago: measuring programming progress by lines of code is like measuring aircraft progress by weight.
The data is brutal. GitClear found an 8x increase in duplicated code blocks between 2022 and 2024. Roughly 41% of new code is now AI-generated, most of it shipped without meaningful review. Teams that push past 40% AI-generated code see rework rates climb 20–25%, and developers who feel 20% faster measure out 19% slower once you count the reviews and bug-fixing. Reward a token budget by lines of code and you're literally paying people to bury you in debt. It's Goodhart's law with a corporate card: the moment output becomes the target, it stops being a measure of anything.
What it actually means
As someone who works in enterprise tech — not as a developer, but close enough to watch these patterns repeat — I've seen this exact movie before. We metered cloud spend and got FinOps teams, showback dashboards, and quarterly fights over reserved instances. Tokens are simply next in line. The same playbook that hides AI's real costs behind a subsidized price is about to hand those costs to engineering managers as a budget to defend.
The healthy versions of Prime's predictions — stipends, estimation, OSS donations — force teams to ask a question they've been able to ignore: is this agent actually worth it? The toxic version rewards whoever generates the most, which is the easiest thing to count and the worst thing to optimize. We've watched everything turn into a metered subscription; AI just became the most expensive one your company runs. The only real question is whether it meters for value or for volume — and the slop cannons are betting on volume.