If it's hard to articulate, that's the evidence — not the weakness

One of the most common critiques of Mixer Mode goes: "if it were real, practitioners could explain it clearly". Polanyi showed seventy years ago that the demand is exactly backwards.

I keep getting this objection in different shapes. From CIOs: "if your senior engineers actually work this way, why can't they write it down?". From academics: "the absence of a clean operational definition suggests you're describing a vibe, not a skill". From other consultants: "you don't have a framework, you have a metaphor people nod at". The objection is reasonable on its face. It's also, I think, exactly the wrong test to apply to mature expertise. Let me explain why — and what to do instead.

Polanyi's Argument in One Line

Michael Polanyi's central claim, in Personal Knowledge (1958) and then tightened in The Tacit Dimension (1966), is that we know more than we can tell. Mature expertise is structurally hard to articulate. That isn't a defect of the expert. It's the signature of expertise. The integrated, embodied, non-verbal layer is where the skill lives. Verbal description is a translation — and like any translation, it loses things on the way.

His canonical example is the cyclist. Ask an expert cyclist how they corner at speed. They produce an incomplete description, sometimes a wrong one — they say they lean into the turn, when biomechanically what's happening is countersteering, with the inside arm pushing the bar in the opposite direction of the turn for a fraction of a second. The cyclist who can't explain countersteering still corners perfectly. The physics teacher who can explain it falls over on the first ride. Articulation and execution are not the same competence, and confusing them is a category mistake.

The demand "if you can do it, you should be able to explain it cleanly" is, in Polanyi's frame, a demand for novice-shaped knowledge. Novices need explicit rules because they don't yet have the integrated layer. Asking experts to produce explicit rules is asking them to perform a regression — to disassemble what's working and present it in a form that fits the asker's vocabulary, not the doer's experience.

The Prediction the Framework Makes

If Mixer Mode is expert mode applied to the whole software-and-agent discipline — Dreyfus stage 5 in the taxonomy Hubert and Stuart Dreyfus laid out in their 1980 Air Force report and then elaborated in Mind Over Machine (1986) — then the prediction is precise: practitioners who run it should find it hard to articulate.

Specifically, they should be able to demonstrate it live. Sit them in front of a real problem with a real agent and you'll see the multi-channel modulation in action — the eye that catches the architectural drift while the hand types the endpoint, the simultaneous read of the test output and the spec implication. But the verbal report afterward — the post-hoc "so what did you just do" — will be partial. Often contradictory. Frequently requiring a metaphor to get close. "It's like mixing audio". "It's like conducting". "It's like driving in heavy traffic — you're tracking everything at once and steering one thing".

That's not a bug. That's the qualitative signature of mature expertise. The framework predicts it, and the prediction has been holding up across the interviews I've run so far. If practitioners articulated it perfectly, that would be the alarming finding — because then what they're running is a simple protocol, not expertise, and the whole pillar would deserve to fall.

What I've Seen in Interviews

The pattern repeats across very different contexts. I sit with a senior engineer who is clearly running multi-channel — you can see it in how they're scanning the agent's output, how they're cutting the spec on the fly, how they're already preparing the rollback while the code is still being written. I ask, "how did you decide that". The answer is almost always some version of I don't know, I just saw it. Sometimes phrased as "it smelled wrong". Sometimes "I felt the architecture was about to bend". Never a clean three-step rule.

Then I offer them a metaphor — usually the mixing console, sometimes the conductor — and the expression on their face changes. They nod, and then they start narrating. The metaphor opened a channel their formal vocabulary didn't have. They're not learning anything new in that moment. They're labeling what they were already doing. That move from "unspeakable" to "speakable via metaphor" is, in Polanyi's terms, the focal awareness catching up with the subsidiary awareness that was already running the show.

The opposite case would worry me, and I look for it deliberately. If a practitioner described their decision process in clean numbered steps, with no hesitation and no metaphor — "first I evaluate the architectural fit, then I assess the test surface, then I decide whether to intervene" — what I'd be hearing is not expertise. I'd be hearing somebody running a checklist they read somewhere, possibly performing for the interviewer. The clean articulation would be a flag, not a confirmation.

How This Changes Senior Interviews

The implication for hiring is direct and uncomfortable, because it invalidates a question almost every senior interview I've sat in has used. Stop asking "walk me through your decision process". You'll get a post-hoc rationalization, the same one they gave the last three interviewers, polished and reordered to sound coherent. It tells you about their narrative skill. It doesn't tell you about their expertise.

Instead, start by asking them to walk through a real decision out loud, as they make it. Hand them a concrete problem from your domain. Let them work it with whatever tools they use — agent, IDE, whiteboard, whatever. Watch the shape of what they do, not the shape of what they say afterward. The think-aloud protocol that cognitive scientists like Ericsson and Simon formalized in the seventies works for a reason: it captures the live operation instead of the retrospective story.

And then ask them, quietly, if they know any metaphors that help them describe how they work when they're in flow. Some will have one ready. Some will think about it and come up with one in the moment. A few will say "I've never thought about it, give me a second" and then produce something striking. The answer to that question tells you more about the depth of their practice than the CV does — because the existence of a usable metaphor means they've already done the work of bringing the subsidiary into focal awareness at least once, and that's the hardest move in the whole skill.

The Implication for the Pipeline

The last move is the most consequential one. If senior expertise is structurally tacit — and Polanyi, Dreyfus, and decades of follow-up work in cognitive science and the sociology of expertise (Collins, Tacit and Explicit Knowledge, 2010) all converge here — then it does not transfer through documentation. You can write the runbook. You can record the demo. You can publish the playbook. None of that will move the integrated, embodied layer from senior to junior. The junior will read the document and produce novice-shaped behavior that looks correct from the outside and breaks the first time the situation is non-canonical.

What actually transfers tacit knowledge is preceptorship. The junior watches the senior decide, live, over months, on real cases. They watch the small adjustments — the one the senior makes silently because the situation just barely turned, the one they skip because the situation just barely didn't. Over time, the junior's own integrated layer starts forming, shaped by hundreds of cases observed in motion. That's how surgical residencies work. That's how master-apprentice transmission worked in every craft before industrialization tried to flatten it.

It's exactly the model Mark Russinovich and Scott Hanselman propose for software engineering in their 2026 piece on AI-era engineering education — and it's exactly what the current pipeline isn't building. The current pipeline is built for explicit-knowledge transfer at scale, because explicit knowledge is cheap to copy. The thing that's becoming scarce is the integrated layer. The thing the pipeline is producing isn't.

Do you have your own metaphor you use to describe how you work when you're in flow? Curious about the ones in circulation that nobody has written down.

#FutureOfWork #Engineering #TechLeadership #Career #Hiring