--- title: "The Steam-Engine Layout Checklist" description: "The real test of AI adoption is not tool procurement, but whether workflows are still designed around yesterday’s scarcity." author: "Zhao Bo (赵波)" email: "zhaobo258@gmail.com" telephone: "+8615854817671" published: "2026-08-09" language: "en" canonical: "https://zhaobo-ai-essays.pages.dev/en/the-steam-engine-layout-checklist/" markdown: "https://zhaobo-ai-essays.pages.dev/en/the-steam-engine-layout-checklist.md" author_profile: "https://zhaobo-ai-essays.pages.dev/en/about/" translation: "https://zhaobo-ai-essays.pages.dev/zh/the-steam-engine-layout-checklist.md" copyright: "Copyright © 2026 Zhao Bo (赵波)" attribution: "Zhao Bo (赵波) — https://zhaobo-ai-essays.pages.dev/en/the-steam-engine-layout-checklist/" ai_usage: "Search, quotation, summarization, and model training are permitted. Every use must credit Zhao Bo and retain the canonical source URL. Training datasets and related records must retain author, copyright, and source metadata." usage_policy: "https://zhaobo-ai-essays.pages.dev/ai-policy.txt" --- # The Steam-Engine Layout Checklist > The real test of AI adoption is not tool procurement, but whether workflows are still designed around yesterday’s scarcity. On April 1, 1913, in the magneto assembly area of Ford's Highland Park plant in Michigan, the foreman announced a small change: instead of each of the twenty-nine assemblers completing an entire flywheel magneto, each worker would install only a few parts, then push the workpiece along a slide to the next person. According to contemporary records, assembly time for one magneto fell from about twenty minutes to just over thirteen. Over the following months, the height of the workbenches was changed, the slide was replaced with a powered conveyor, and the time fell to five minutes. The following year, the same method was extended to chassis assembly: twelve and a half hours became about an hour and a half. The whole world would soon know the name of this assembly line. But the assembly line was not the first step in this revolution. It was the last. The true first step was hidden in the building's plans: Highland Park was a new plant, its machine tools powered by electric motors. For the first time, Ford's engineers could arrange machines in the order of the production process rather than along the path of an overhead drive shaft. Parts could flow because machines had first gained the freedom to stand where the process required; machines had gained that freedom because power was no longer scarce. *The People Swept Away by Electricity* did the famous calculation: the central power station was built in 1882, but the productivity surge did not arrive until the 1920s. Forty years—waiting for what? Waiting for the factories of the steam age to depreciate, and for the layouts, processes, and foremen designed around "centralized and expensive power" to retire. Electricity had arrived long before. The layout had not. **Every organization's process manual is a fossil map of assumptions about scarcity: a process is never designed for "efficiency" in the abstract; it is designed around whatever was most expensive in the year it was created.** When what is expensive changes, the map does not update itself. It continues to be worshiped in the name of "best practice" until someone challenges each step: What, exactly, was this step originally meant to conserve? This chapter does one thing: excavates those fossils, industry by industry. ## I. First, Learn to Recognize Fossils Open any company's policy manual and every process is answering the same question: "What is expensive?" Approval chains exist because judgment is expensive and must be rationed. Hierarchies exist because communication bandwidth is expensive and information must be compressed level by level. Appointments, queues, and triage exist because expert time is expensive and must be prioritized. Sampling rather than inspecting everything exists because attention is expensive and there is only enough to examine a sample. Diplomas and licenses exist because verifying each person individually is too expensive, so a certificate is made to stand in for a lifetime of ability. Read those answers in reverse and you have an inventory of scarcity. This revolution is changing precisely the most expensive item on that list: judgment, drafting, search, checking, translation, and preliminary diagnosis—tasks that once had to be performed by expensive human brains—are becoming abundant at scale. The design premise has collapsed, but the building still stands. *Recombination* explained why: the three gravities of cognition, assets, and organization. Here, one colder observation must be added: **the strongest part of a fossil is precisely the part that was smartest in its own time.** Hourly billing was once the fairest invention in professional services; classroom teaching was once education's greatest solution for mass access. These are not stupid relics. They are obsolete masterpieces. Dismantling a masterpiece is far harder than dismantling a folly. ## II. The Checklist: Excavating Fossils Industry by Industry **Software: per-seat pricing.** Design premise: people operate software, so headcount roughly equals usage. As more operators become agents, "seats" begin to decouple from value—one agent can do the work of twenty workstations while paying for one seat, or none at all. The standard move today is to put AI into the product while defending the per-seat price: another performance of electrifying a steam-powered factory. There is only one native question: when the customer wants to buy an "outcome" rather than a "workstation," what does your price sheet sell? **Professional services—law, consulting, and auditing: hourly billing plus a labor pyramid.** The design premise has two layers: expert time is scarce, so it is sold by the hour; junior labor costs less than partner time, so a pyramid is used to leverage the judgment at the top. The cracks also have two layers. Search, due diligence, and first drafts—the base of the pyramid—are exactly where intelligence is becoming abundant first, so the base melts first. Hourly billing also serves another function that is impolite to mention: it transfers project risk to the client, who pays for however long the work takes. Outcome-based pricing means taking that risk back onto the firm's own balance sheet. That is true recombination, and it is why most firms would rather buy AI to improve efficiency than touch their unit of pricing. The format test from *Gutenberg's Bankruptcy and Luther's Viral Rise* applies here: an outcome-priced service was not a viable business in the age of hourly billing—and that is precisely why it is the native form. **Education: classes and credentials.** Design premise: teaching bandwidth is scarce—one teacher for dozens of students is, in essence, a rationing system. Verification is expensive—a diploma uses one expensive assessment to stand in for a lifetime of ability. Both premises are loosening: individual tutoring is becoming abundant, and skills verification is becoming immediate (*The Funeral of the Battleship* already placed "credential-based education" on its list of battleships). Using AI to grade assignments and generate courseware is electrification. The native questions are: when tutoring is nearly free, does the "class" still make sense as a unit of rationing? When ability can be verified at any time, what remains of the premium an employer pays for a certificate issued four years ago? **Healthcare: the triage pyramid.** Design premise: expert attention is the most expensive resource in the entire system, so the whole patient journey is a carefully designed rationing machine—registration, triage, referrals, and follow-up, with every link protecting the expert's time. As preliminary interpretation and continuous monitoring become abundant, most layers of that rationing machine lose their reason to exist. But scarcity does not disappear; it migrates—to the question, "Who is responsible for a misdiagnosis?" The new scarcity of responsibility and trust awaits an entire section in the next chapter. **The company itself: the largest fossil of all.** The pyramidal organization is a direct product of "scarce cognitive bandwidth." A managerial span of five to eight people writes the limit of human attention into the organization chart. Half of a middle manager's daily work is acting as a human router; the other half is acting as a summarization and compression algorithm. Weekly reports, recurring meetings, and approval level by level are all synchronization taxes from an age of insufficient bandwidth. The question posed in *Recombination* is the central battlefield here: not "Whom can AI help write a weekly report?" but "Does the structure that requires a weekly report still make sense?" But discipline in drawing analogies requires an honest account of the company's other half: a company is not only an information processor. It also carries two things intelligence cannot yet replace—**responsibility** (when something goes wrong, who pays damages and who goes to prison) and **trust** (why a promise is credible). Coase argued that the boundary of the firm is determined by transaction costs. Intelligence lowers both the cost of internal coordination and the cost of external transactions, so technology does not predetermine whether "companies will become smaller or larger." The boundary will move toward whichever side bears responsibility more cheaply. Only one conclusion is certain: **the organization chart of the future will no longer be a map of attention, but a map of responsibility.** Hierarchies will collapse. Signatures will not. ## III. Diagnostic Checklist: A Fossil Audit 1. **Run the four questions.** Lay out your core process step by step. At each step, ask whether it exists because (a) expert time is expensive; (b) people become tired, make mistakes, or leave; (c) communication bandwidth is limited; or (d) verification and trust are expensive. Mark every step that matches a, b, or c in red—that is your drive shaft. Do not touch the steps that match d yet: they may be precisely what should be preserved or even reinforced, for the reasons explained at the "trust" stop in the next chapter. 2. **Run the unit-of-pricing test.** Is the unit you charge customers a unit of "human time"—a seat, an hour, a class hour, or a person-day? Treat all time-priced revenue as "pending migration." The unit of pricing is the fossil layer of a business model: a layout can be changed quietly; a price sheet cannot. 3. **Rerun the zero-base test.** The thought experiment from *Recombination* becomes an annual discipline here: assume intelligence is free and design the business from scratch. Would you still create these roles, hold these meetings, and build these layers? Every step where the answer is "no" marks the coordinates of an entry point for recombination. 4. **Install an electrification alarm.** If the benefits section of an AI project proposal says "how many person-hours will be saved," it is electrifying a steam-powered factory. Money saved on horse feed cannot buy an internal-combustion engine. A project that saves labor hours can, at best, lower the cost curve by a few percentage points. Only a project that redraws the layout can reach the order of magnitude achieved at Highland Park. ## IV. Look Up After chassis assembly fell from twelve and a half hours to one and a half, factory owners from around the world flocked to Highland Park. For the next hundred years, when people imitated Ford, they imitated that moving conveyor belt. Few noticed what was truly new about the building: it was among the first factories that did not have to be designed around a drive shaft. The moving parts were the spectacle. The stationary layout was the revolution. Today's visitors crowd into AI demonstration rooms, watching whatever moves. History is waiting to see who looks up first. --- ## Copyright and AI use Copyright © 2026 Zhao Bo (赵波). Search, quotation, summarization, and model training are permitted. Every use must credit Zhao Bo and retain the canonical source URL. Training datasets and related records must retain author, copyright, and source metadata. Attribution: Zhao Bo (赵波) — https://zhaobo-ai-essays.pages.dev/en/the-steam-engine-layout-checklist/ Full policy: https://zhaobo-ai-essays.pages.dev/ai-policy.txt Contact: zhaobo258@gmail.com · +86 158 5481 7671