Field notes · essay
KCS isn't dead. Its assumptions are.
Periodically someone announces that KCS is dead. The argument runs roughly like this: AI writes the articles now, so a methodology built around agents writing articles is finished. I understand the impulse. I think it gets the story exactly backward.
KCS didn't fail. It succeeded so completely that the world it was built for is gone.
Look at what it argued for, back when the arguments were hard. Capture knowledge as a byproduct of solving problems, in the words of the people who had them. Create content because demand proved it was needed, not because someone predicted it would be. Let the people doing the work own the knowledge instead of handing it off to be written up at arm's length. Improve an answer every time someone uses it. Twenty-five years ago, each of those was a fight. Today nobody serious argues the other side, which is what winning looks like for a methodology: its principles stop sounding like opinions.
And the corpora those programs built (resolved issues, in the customer's own language, linked back to the cases that produced them) turn out to be precisely the material an AI system needs to ground an answer. KCS programs spent years building the context layer before anyone had a name for it.
The stewards seem to agree the scope has grown. In April 2026, at Service Innovation Summit 2026, the Consortium for Service Innovation renamed the methodology: Knowledge-Centered Service became Knowledge-Centered Success. Their announcement calls KCS "an operating model that makes AI viable at scale." I agree, and I'd attach the condition this essay is about: it does that once both of its founding assumptions are retired on purpose.
Principles outlive their assumptions
Every methodology makes two kinds of claims. Principles are about the work: what's true of knowledge and how it gets made. Assumptions are about the environment the method was born into: who does the labor, what the tools can do, who's on the other end. Assumptions are load-bearing, which makes them easy to mistake for the point. When the environment moves, assumptions expire. Principles don't.
KCS carried two load-bearing assumptions. Humans write the knowledge. Humans read it. Both held for twenty-five years. Neither holds now. The founding essay argues each in full; here is the short version, and then the part that matters on Monday morning.
Humans write: expired
The hardest thing KCS ever asked was for an agent, mid-queue, to stop and write. That's the capture tax, and it's the joint where most struggling programs I've seen broke: leadership says knowledge matters, then measures the agent on throughput, and the agent does the math.
The tax is gone. A resolved case already contains the problem, the environment, the troubleshooting path, and the fix. AI drafts a structured article from it in seconds, with a link back to the source case. The agent's part shrinks to one judgment, about thirty seconds long: is this one worth keeping? The human work doesn't disappear. It moves to curation: review the draft, fix what's wrong, approve or reject at the gate. How much the curator changed becomes a live benchmark of drafting quality, the edit-distance pattern Michael Iantosca described.
The principle survived intact. Knowledge is still a byproduct of solving. Only the typing changed hands.
Humans read: expired
Even in the good years, people rarely read articles the way the templates imagined; they skimmed a paragraph mid-crisis, after search handed them the page. Now the first reader is usually a machine: the agent-assist panel, the help-center bot, the copilot a customer wired into their own service desk, and soon customers' own AI agents deciding on their principal's behalf. The machine doesn't visit the article. It retrieves pieces of it as context for an answer a person reads instead.
So the quality bar moved. "Sufficient to solve" was written for a person on the page. For a machine it means sufficient to ground: a section that survives being retrieved alone, explicit applicability (which product, which version, which plan), a stable identifier an answer can cite tomorrow, and a freshness signal that says whether it's still true. Those are the four clauses of Machine-Ready Knowledge. Each one extends something KCS already believed. None of them repudiates it.
Sort the practices, keep the principles
"Is KCS dead?" is the wrong question for a program to ask. The useful one is narrower, and you can ask it of every practice you run: which assumption does this rest on?
- Rests on "humans write": the authoring step at case close, coaching aimed at writing skill, quality reviews scored on prose. Rebuild these around the flag and the gate. Coach curators' judgment, and read edit distance as the coaching signal.
- Rests on "humans read": templates tuned for skimming, search tuned for people typing keywords, style rules about scannability. Rebuild these around retrieval: self-contained sections, declared applicability, stable identifiers, freshness.
- Rests on neither: capture in the flow of work, content created where demand proves the need, ownership by the people doing the work, answers that improve through use, the loop between work and knowledge. Keep every one exactly as it is. That's the soul.
Notice which list is longest.
Knowledge management is becoming context management
Knowledge management optimized the article for a visitor. Context management governs the passage a machine assembles into an answer, and that means governing the record those passages come from: who can write to it, how changes get reviewed, when content expires. The upkeep work everyone agreed was important and nobody staffed is now the load-bearing wall. The rule that holds it up is one sentence: nothing probabilistic writes to the record without passing a human gate.
The stakes rise with polish. A clumsy chatbot could be wrong charmingly. A fluent answer engine that cites your own stale article to a customer's agent, mid-decision, burns trust you don't get back: the precision paradox. Retrieval can't rescue an ungoverned corpus, and no model outruns the context it's given.
Which is why practitioners with years of KCS behind them aren't behind at all. They've been doing context management since before AI gave it a name. The discipline they learned is the one the AI era runs on; it just has a new kind of writer to govern and a new kind of reader to serve.
KCS isn't dead. Its assumptions are. That's the best thing that can happen to a methodology: its principles get to outlive the world that produced them.
The full argument, including the three governed loops that grow out of KCS's two, is the founding essay. The four clauses a machine reader needs are published, free, at machinereadyknowledge.com.
KCS® is a service mark of the Consortium for Service Innovation™. This essay reflects an independent, KCS-informed methodology and is not affiliated with or endorsed by the Consortium. The edit-distance benchmark is credited to Michael Iantosca and draws on his published papers at thinkingdocumentation.com/downloads.