SECI vs. KCS: two altitudes of the same problem

Placeholder. A real draft might work through where KCS’s Externalization/ Internalization loop lines up with SECI, and where KCS structurally under-serves Socialization — the tacit-to-tacit transfer that never gets written down.

The correspondence

KCS’s central discipline — capture knowledge in the workflow, at the moment of demand, then let it mature through reuse — is Externalization and Internalization running in a tight loop. An engineer solving a case holds tacit knowledge; writing the article as they solve it is Externalization. The next engineer reusing and improving that article, absorbing it into their own practice, is Internalization. KCS’s content-maturity model — draft capture growing into validated, reusable content through use — is Nonaka’s spiral made operational and measurable.

Where AI tooling actually sits

Look at what AI capture tools mechanically do: summarize a case, draft an article from a resolved ticket, cluster similar tickets. That’s operating on text that’s already explicit — the case notes, the transcript, the resolution someone already typed. That’s Combination: explicit knowledge reorganized into other explicit knowledge. The genuine Externalization moment — the judgment call, the thing an engineer knew but didn’t type — already happened before the AI touched anything, or it never got typed and the AI has nothing to work with.

Decompose an actual solve-loop interaction and this becomes clearer. A case comes in. Two engineers swarm it — Socialization, tacit-to-tacit, before anything’s written. One writes the article — Externalization. A different engineer later finds it during reuse, comparing it against related articles — Combination. Repeated reuse folds the pattern into their own judgment — Internalization. All four modes, compressed into what KCS just calls “capture” and “reuse.” The double loop was always running the full spiral; KCS just never named all four steps, because it only needed to instrument the two that produce artifacts.

AI tooling, because it only has access to what’s already text, can only ever touch the Combination slice of that spiral. It’s not that AI misses Externalization by accident — it’s structurally blind to Socialization entirely, because the swarming conversation is the part least likely to ever become text in the first place.

The blind spot is the room, not a gap

That blind spot isn’t empty. It’s where the actual cognition happens.

Nonaka calls the space that holds this kind of knowledge creation Ba — a shared space, physical or otherwise, that exists specifically to hold the act itself. The originating Ba is unstructured, often unproductive-looking time: two engineers at a whiteboard, a swarm call where half the value is in the silences and interruptions, someone saying “walk me through what you were thinking” and watching the hesitation as much as the words. That hesitation is data — but it’s relational data. It only exists between two minds in the same space at the same time. It was never sitting anywhere to be extracted from.

That’s a different kind of blocked than “not enough training data yet.” Externalization is representational — one mind converts its own tacit knowledge into a symbol that then exists independently and can be operated on. Socialization is relational — the knowing happens between people, constituted by the interaction itself, and produces no artifact until someone later externalizes it. AI works on artifacts. It is downstream of the place where this cognition happens, not present for it.

That’s the actual crux of “AI isn’t here to take your job,” stated more precisely than the slogan usually manages: it isn’t a labor-economics claim, it’s an epistemological one. The valuable cognition — judgment calibrated in real time against another person’s pushback, trust built by being visibly wrong together and correcting — has no substrate other than humans in a room, or on a call, with each other. AI can compress and distribute what comes out of that afterward. It cannot be in it.

This reframes what “human in the loop” ought to mean. The common usage treats humans as a QA checkpoint on AI output — approve, edit, reject. The Ba argument says humans aren’t checking the loop; they’re generating the substance the loop runs on. Coaching, in KCS terms, isn’t a soft-skill add-on to the real technical work. It’s the discipline of deliberately running Ba — creating and protecting the conditions where that relational cognition can happen, because nothing else in the system can produce it.

The human and their purpose is the point. Not a placeholder until the tooling catches up — the irreducible center the tooling exists to serve.