Here is the machine's catastrophic structural weakness, stated plainly: it can build anything. It has no idea what to build.
Give an AI system a well-specified task and it executes with extraordinary capability. The software gets written. The analysis gets run. The document gets drafted. The market research gets synthesized. The report gets formatted. The presentation gets built. All of this happens faster and more cheaply than a human can do it, and with quality that rivals or exceeds specialist human output.
Now ask it: should we build this software? For whom? What problem does it solve that isn't already solved? Which interpretation of the requirements is the one the client actually means? Which version of this analysis matters to the people who will act on it? Which risk is worth taking given Q4 cash flow constraints that aren't in any document?
The machine processes publicly available information and pattern-matches against its training data. It cannot know what is true about your specific client, your specific supplier, your specific regulatory environment, your specific organization at this specific moment. That knowledge lives in people who have operated in those contexts over time. It is not derivable from publicly available data. It is not in the training set. It is, in the most precise sense, irreplaceable.
The Architect vs. The Analyst
The Analyst executes: runs analyses, produces outputs, processes information, formats deliverables. AI now performs Analyst work at near-zero marginal cost. The Analyst role has been commoditized.
The Architect defines the problem: establishes success criteria, applies domain judgment about which outputs are meaningful, makes the decisions that require accumulated context, and takes responsibility for whether the machine is pointed at the right target. The Architect role has increased in value as Analyst work has been automated.
Most professionals have been operating as Analysts. The transition required is not a skill upgrade — it is a role migration.
This distinction matters because it changes what "career development" means in the AI economy. The traditional path — develop execution skills, build a portfolio of execution outputs, get promoted for execution excellence — is the Analyst path. That path terminates in a repricing event.
The Architect path: develop deep domain expertise in a specific field, build judgment through years of operating in that field, learn how to direct AI execution within that field, build the ability to explain AI outputs in terms that stakeholders without domain knowledge can act on. That path has become more valuable as the Analyst path has been automated.
The Domain Translator
The Domain Translator — Sterling's term for the specific role that has increased most in value — is someone who possesses irreplaceable knowledge of a specific field and can function as the bridge between that domain and the AI execution layer.
What makes domain expertise irreplaceable is its specificity. Not general knowledge about an industry — that's in the AI's training data. The specific knowledge of: which supplier delivers reliably in Q1 and becomes unreliable in Q4 because their primary manufacturer shuts down for inventory; which regulatory interpretation the enforcement division actually applies versus the one that's technically on the books; which version of a client's stated requirement is the one they mean versus the one they wrote down; which project risk is the kind that sinks things quietly versus the kind that's loud but manageable.
This knowledge is not in any document. It is in the person who has operated in that specific context, with those specific people, in those specific conditions. No training set captures it. No model derives it from general patterns. It is accumulated through experience that cannot be replicated by a system that has never experienced anything.
The Paradox of Intent says: the more capable the execution layer becomes, the more valuable this knowledge is. Not because humans are inherently irreplaceable — they aren't, at the execution layer. Because the execution layer, left without direction, produces nothing of value. You need the map to use the vehicle.
How to Price Domain Expertise
The domain expertise was previously bundled with execution work. The execution work was visible, measurable, billable by the hour. The domain expertise was invisible — embedded in the judgment about which execution work to do, present in the results but never explicitly priced.
When AI decouples execution from expertise, the expertise that was embedded in the execution has to be priced explicitly. The market does not automatically price it — it has no mechanism for doing so. The professional who has unbundled the expertise from the execution has to name it, demonstrate it, and charge for it as the primary value delivered.
This is what The Skill Bankruptcy calls Anti-Commodity Positioning: moving from billing for execution hours to pricing for domain judgment outcomes. The rate per hour is not the relevant metric. The value of the decision informed by the domain expertise is the relevant metric.
AI's structural limitation: it can execute any task it is directed toward, but cannot determine which task matters. The irreplaceable elements: knowing why the software is being built, which supplier is reliable in Q1 but unreliable in Q4, which regulatory interpretation has teeth, what the client actually needs vs. what they think they want. These require accumulated domain-specific experience that cannot be derived from publicly available training data.
The Analyst executes tasks — AI has commoditized this at near-zero cost. The Architect defines problems, establishes criteria, applies domain judgment, and makes decisions requiring accumulated context. The Architect role has increased in value as Analyst work has been automated. Most professionals have been on the Analyst path — the transition required is a role migration, not a skill upgrade.
Sterling's framework: as AI execution capability increases, the value of human intent increases proportionally. The more the machine can do, the more important it becomes to know what to tell it to do. The scarce resource has migrated from execution to direction. This is not a reassurance about human uniqueness — it is a precise map of where value lives in the AI economy.
Sterling's test: can your primary work output be produced by an AI given a well-crafted prompt and publicly available information? If yes, that output has been commoditized. Follow-up: what judgment would a human need to provide to make the AI's output actually useful in your specific context? That judgment is your non-automatable asset.
Necessary but not sufficient. The survivable position is domain expertise plus AI architecture capability. Domain expertise without AI architecture is a library without a search engine. AI architecture without domain expertise is a machine without a map. The Domain Translator combines both: deep domain knowledge that directs AI execution, plus the capability to build and direct the AI execution layer.
