Context debt
Anytime someone on your go-to-market team uses a language model, it needs a slice of your strategy before it can be useful: company, product, buyers, positioning, plays, proof points. That context now lives somewhere - a prompt doc, a Claude project, a tool's settings. And every one of those places is a debt you carry: when strategy moves, someone re-propagates the change by hand, everywhere, or the output quietly goes wrong.
The mechanics
The debt compounds at every strategic update. A launch ships, a competitor emerges, a win/loss insight re-cuts the ICP - and every Claude project, every automation prompt, every enablement doc, every point-tool configuration has to be updated to match. The burden grows along two axes at the same time: the number of surfaces you have spread context across, and the speed at which your strategy changes. Neither axis slows down as AI adoption grows. Both accelerate.
A bigger context document does not retire the debt. It concentrates it: one heroic file that goes stale as a unit, feeding every surface the same aging copy with the same confidence.
The version of this I lived
I spent two years paying this down by hand. I didn't want a billion more ways to send AI emails; I wanted a centralized brain that could feed every place our go-to-market touched a model. So we built - vector stores wired into our automation, shared Claude projects for every campaign and event, DIY context everywhere.
Every build increased the debt. We would run an enablement session on new packaging and walk out with a long, spotty list of AI prompts and context stores we now had to update by hand to keep up with our own strategy. One step forward, one step back. The lesson was not that the tools were bad. It was that a pile of context stores is not a context system - and without the system, every tool you add is one more place the debt accrues.
The five symptoms
Any three of these and the diagnosis lands:
I. "Our AI outputs feel generic, off-brand, inconsistent." The strategy isn't reaching the model uniformly.
II. "Different teams are using AI in totally different ways." Individual pockets of leverage, no shared system underneath - each person's private context, private wins, private staleness.
III. "We launched a thing and it took weeks for sales to use the new messaging." Propagation is human-paced.
IV. "We've bought five AI tools and they all need their own context configuration." You are paying the debt N times, where N is the number of surfaces.
V. "When someone leaves, knowledge walks with them." The context lives in heads, not systems.
The crossover
Here is the asymmetry that makes this worth fixing rather than managing. Context debt compounds linearly: every strategic change costs another round of manual propagation, forever, and the cost rises with every surface you add. A shared, curated context system compounds geometrically: every piece of context added makes every person and every agent that draws on it better, at once. Those two curves cross.
You reach the crossover two ways, and you need both: centralization (one place the context lives, not N copies paying the debt N times) and freshness (the system stays current, so the output stays trustworthy). Miss either and you are back in debt - centralized-and-stale is just a bigger document; fresh-but-scattered is the two years I lived.
What sits past the crossover - the layers, the owners, and the mechanisms that keep the system fresh - is the main piece: Go-to-Market Context Is a System, Not a Document.
The system that retires the debt.
The layers, the owners, and the freshness mechanisms - the full argument this note hangs off - live in the main piece.