The industry moved from per-seat licensing to consumption pricing, and bills started drifting upward. Gartner has criticized vendors for a lack of transparency here, and teams report jumps from twenty or a hundred dollars to thousands per developer per month.
Below: where that spend comes from and what actually helps.
Why the bill grows quietly
It comes down to how agents work. A chat message is one call. An agent working a task makes dozens, and each call drags everything before it into the context.
That gives three quiet sources of spend:
- Context piles up. Step twenty costs more than step one because it carries the history of every earlier step.
- Retries. The agent makes a mistake, reads the error, tries again. Every loop is billed.
- Background runs. An agent in CI fires on every commit, so spend follows your team's activity, not your decisions.
The most annoying line item
One widely repeated estimate says a substantial share of tokens goes to fixing bugs the model itself wrote.
Think about that: you pay twice, first for the problem to appear and then to find and remove it. The practical takeaway saves more than any micro-optimization. The sharper the task going in, the less you pay coming out.
What lowers the bill
Split models by task. Keep the flagship for work that ties a lot of context together. Parsing logs, rewriting text and generating boilerplate run fine on a smaller model.
Cap the answer length. Output costs several times more than input. Asking for "code only, no explanation" trims the bill noticeably and breaks nothing.
Do not put a whole file in the context when you need one method. Input is cheaper than output, but not free, and on agentic runs it compounds.
Set the ceiling up front. A monthly cap with an alert at seventy percent settles the problem before it turns unpleasant. In the codexkey dashboard the cap and its thresholds are configurable per user and per team.
Seeing where the money goes
A total tells you nothing until you can break it down. You need to know which key, which project and which model produced the spend.
We report usage per key, and sub-organizations let you move teams and projects onto their own budgets. There is nothing to reconcile by hand at month end.
The short version
Agent spend is not a fixed line. It depends on how you frame tasks and which models you pick. A ceiling, a per-key breakdown and sensible model choice remove most of the problem.