06-reference

commoncog deming paradox

Tue Apr 14 2026 20:00:00 GMT-0400 (Eastern Daylight Time) ·reference ·source: Commoncog ·by Cedric Chin

“The Deming Paradox: Operationally Rigorous Companies Aren’t Very Nice Places to Work” — @CedricChin

Why this is in the vault

Chin names a tension that sits directly under the MAC framework and the agent-deployer role: the same rigor that makes operations excellent tends to grind the humans executing inside it. If RDCO is selling MAC as “SPC for AI models,” we inherit this paradox by default — so the vault needs an honest entry on it.

The core argument (paraphrased)

Deming’s data tools (SPC, process behaviour charts) have a documented track record of producing operational excellence — Amazon, Koch, Toyota, Honda, Sony. They also have a documented track record of “sucking the soul out of everyone” who works inside them. That’s the paradox.

The odd part: Deming himself espoused a deeply humanistic philosophy. He argued against ranking employees, numerical quotas, grading, and the assumption that problems are caused by workers rather than the system. He wrote that “it is wrong to suppose that if you can’t measure it, you can’t manage it.” His “System of Profound Knowledge” is meant to replace management-by-fear with management-by-understanding-variation.

So how do the tools end up producing crushing workplaces? Two possibilities, and Chin isn’t sure which dominates:

  1. Ignoring the humanistic half. Companies adopt the charts and process discipline but skip the “treat workers with joy and respect” part. The Koch/Dubose barge story is the tell: run charts posted in every skipper’s cabin, each boat turned into a P&L, skippers competing against each other, “the people who couldn’t support that, well, most of them were let go.” Operational excellence via visibility plus pressure plus replacement.
  2. The tools themselves generate the pressure. Making variation legible makes underperformance legible. Once you can see who’s on the wrong side of the run chart, the humane response (“the system produced this, not the person”) is hard to hold onto in practice.

Chin’s open question: has any company ever implemented the full Deming system — rigor AND humanism — and made both halves work? He suspects the answer may live in Japanese auto history, but he doesn’t know yet. Koch got the rigor and got the results, and also got a Senate oil-theft investigation in 1988 that Leonard traces directly to internal continuous-improvement pressure.

The takeaway Chin leaves us with: Deming’s ideas are tools, and tools can be misused. Operational rigour without the humanistic frame is a loaded gun pointed at your team.

Mapping against Ray Data Co

mapping_strength = strong. This article is the uncomfortable counterweight to the first-principles piece. If that essay is why MAC works, this one is why MAC can hurt.

1. MAC makes quality legible, which makes people legible. The 3×6 MAC matrix turns data model behavior into run-chart-equivalent signals — stop/pause/go tiers on scope × basis. That’s exactly the Deming move: make variation visible, differentiate routine from exceptional. But the Koch/Dubose pattern warns us that once quality is visible at the cell level, so is the operator behind the cell. A dashboard that shows “this team’s models keep hitting Stop severity” is a performance management tool whether we intend it or not. The coaching curriculum needs to explicitly address this or clients will weaponize MAC against their own data teams.

2. The agent-deployer role inherits the paradox. Per 2026-04-14-levie-agent-deployer-role-jd, the agent-deployer runs evals, instruments workflows, manages quality gates. That’s the modern SPC operator — and per Chin, SPC operators working in rigorous companies tend not to be happy. If RDCO pitches agent-deployer as the hot new career, we’re also pitching people into the role that, historically, has the highest human cost in operationally excellent firms. Worth being honest about in any JD we help write.

3. State-ownership is a partial mitigation. ../04-tooling/rdco-state-ownership-architecture argues the client owns the vault, the skills, the state. That matters here because one of Deming’s core humanistic claims is that the system produces outcomes, not the individual worker. If the operator owns the system (vault + skills), they have standing to change it when it produces bad outcomes. Contrast with Koch’s barge skippers, who owned their P&L but not the meta-rules. Framing state-ownership as operator agency over the system — not just data portability — makes RDCO’s posture more Deming-humanistic and less Deming-Koch.

4. The phData/MG comparison needs a humanism axis. When we position against phData (staff aug + managed services) vs MG (McKinsey playbook imports), we usually argue on cost, speed, and state-retention. Chin suggests a fourth axis: what does this engagement do to the client’s team dynamics? A phData staff-aug engagement is low-rigor and low-pressure; an MG playbook is high-rigor and high-pressure. RDCO’s pitch should be high-rigor, medium-pressure — explicitly softer than the Koch mode because the MAC frame is paired with coaching, not performance reviews. This should show up in sales decks.

5. Honest marketing, not operational-rigour porn. Chin’s essay is a data-literacy author publicly admitting his methods have a body count. That’s a posture worth copying. The Sanity Check newsletter and any RDCO sales material should not sell MAC as pure upside. Pair every “operational excellence” claim with the humanism guardrail — explicitly. Helps inoculate against the failure mode where a client adopts MAC, uses it to fire three data engineers in Q1, and then blames the framework.

One unresolved question for RDCO. Chin asks whether any company has done Deming end-to-end — both halves. For our narrower scope (MAC + agent-deployer + state-ownership), the equivalent question is: has any consulting engagement produced durable data quality discipline without producing a worse place to work? If we can’t point to one, we’re flying blind on the humanism half, same as everyone else.