The Context Tax: The Invisible Cost of AI That Can't See the Work

AI vendors quote per-seat pricing. The number on the invoice is the visible cost. The cost nobody invoices is the one paid in seconds — every time an employee has to break what they're doing, summarize it in prose, and feed that prose to a chat window before the model can help. On most deployments it is larger than the license fee. Dataken calls it the Context Tax.

The tax shows up in four places. Re-explaining intent — the model has no memory of the screen the user was just looking at, so context gets retyped every session. Prompt drift — two analysts ask the same question with different words and get different answers, so output quality becomes a function of phrasing. Tool dormancy — the seat exists, but the user can't remember what to ask, so the license quietly becomes shelfware. Adoption decay — month two's usage is half of month one's, because friction outlasts novelty.

Each of those is a small tax. Multiplied across a function, then a quarter, then a workforce, it's the line item that explains why the ROI in the business case never shows up in the P&L.

Better prompt templates don't fix this. They reduce the per-instance cost slightly and leave the structural problem in place — namely, that the human is being asked to do work the agent should have done by reading the screen.

OLi removes the tax on workflows where it sits because the context is observed, not typed. The user keeps working. The agent already knows what they're doing. The seconds that would have been paid to the Context Tax stay in the user's day.