AI quota inflation is no token effort. It's baked in

The problem, bluntly put

Rupert Goodwins argues in a new opinion piece for The Register that the way large language models are billed — by tokens in and tokens out — practically guarantees creeping price pressure. He dubs the phenomenon TIBS, “token incremental burn syndrome,” and warns we may be at the start of something that’s hard to stop. Simple to count? Sure. Simple to make fair? Not at all.

Tokens: neat, wrong, inevitable

The piece lays out the core gripe: tokens are an easy metering unit, like paying programmers per keystroke, but they don’t map to useful work or value. Benchmarks can measure tokens per second or output/input ratios, but that doesn’t tell you how much business value you actually got. It’s a tidy billing story for vendors — messy for customers. Want efficiency or outcomes? Good luck tying those to a token invoice.

Lock‑in and the frog‑boil effect

It has been reported that vendors have an addiction to subscription models, and Goodwins paints a pretty bleak picture: deskilled teams, baked‑in dependencies on a single code‑generation chain, and migration that’s prohibitively painful. Call it vendor lock‑in, call it inertia, call it what you like — the emotional punch here is real. Once an org relies on a specific LLM pipeline, leaving feels like jumping from a slow cooker mid‑simmer.

Why this matters now

The upshot: token‑based billing plus subscription pressure creates a plausible recipe for inflationary creep in AI costs. If the industry needs outsized returns to justify massive AI investment, vendors will look for levers — and tokens are the easiest pull. So what should you watch? AI‑heavy CI/CD bills, changing subscription terms, and any metric that rewards churn over outcomes. Can the market invent a better way to measure value? It’d be nice. But don’t bet your budget on waiting for miracles.

Sources: The Register