Context debt: the cost you are already paying

Gartner coined a term recently that I think is going to stick: context debt. They define it as the loss of understanding of not only what happens in a process, but why.

Every company carries this debt already. When a new hire takes six months to become useful, that is context debt. The facts were all there in the wiki and the CRM. What took six months was the why: which customers are fragile, which claims legal actually approved, why the pricing exception exists, what the last rep promised. None of that is written down anywhere a system can find it.

AI did not create the debt. It sent a new collector.

The interest payments

Here is what the interest looks like day to day. A rep asks a model for an account brief and gets back something a stranger could have written, because the model is a stranger. A marketer gets copy that has to be rewritten line by line against the approved claims, because the model never saw the approved claims. Someone in finance explains the revenue definition to a chatbot for the fortieth time this quarter.

Each instance costs minutes. Multiply by every person, every prompt, every day, and you get the number that shows up in the surveys: nearly every company is using AI, and almost none of them can point to the productivity. The models are fine. They are paying interest on the missing context, one prompt at a time.

Why another tool will not fix it

The instinct is to buy a better tool. The new tool arrives knowing exactly as much about your business as the last one: nothing. So each team starts teaching it from scratch, in prompts, in pasted documents, in custom instructions that live on one person's laptop.

Now the debt is spread across a dozen tools instead of one, each with its own partial, private copy of how the business works. The industry has started calling these context islands. They do not pay the debt down. They just make it harder to audit.

Paying it down

There is one fix, and it is structural. Put the knowledge, the rules, the language, and the history in one governed place, and feed that place to whatever AI you use. Fund the context once, and every tool, every team, and every prompt draws on the same balance.

That is the entire argument for a context layer. It converts a debt that compounds against you into an asset that compounds for you: every account, every call, every decision that flows in makes the next piece of work better.

The debt is already on your books. The only question is whether you keep paying interest or start paying principal.