Where the cohort comes from

Use this note as a worked instantiation, not an answer to repeat. The transferable question: when a capability is emerging faster than any institution can form workers in it end-to-end, what are the plausible upstream pipelines for the eventual specialist cohort, and how would you judge in advance which is likeliest to dominate? This note's instantiation: for AI-era auditable automation of regulated reporting, four candidate pipelines -- formal-methods engineers who pick up the tacit supervisory half, regulatory experts who learn to build, regtech vendors as accidental finishing schools, and supervisor alumni -- each compresses a different half of the composite and leaves a different tacit residue uncovered. The note favors the regulatory-experts-learn-to-build path on one argument: error asymmetry, since a domain expert's bad code fails visibly while an engineer's bad regulatory judgment fails invisibly until the supervisory letter arrives. Apply the question to your own emerging-capability domain: enumerate the plausible upstream pipelines, name which half of the composite each compresses versus leaves tacit, and identify which pipeline's failures are cheapest to catch as a proxy for which hiring will actually favor. Produce a ranked list of candidate pipelines with the failure-cost argument for each, checked against real hiring examples where you have them.

Companion to The candidate produced somewhere else. That note claims one capability in current senior specs has no apprenticeship system yet: AI-era automation of regulatory reporting under supervisory evidence requirements. This one speculates about where the eventual cohort comes from. Speculation is the operative word; check this against hiring patterns in a few years.

Four paths look plausible.

Formal-methods engineers. Regulation is quietly becoming a formal artifact: reporting taxonomies are typed schemas, validation rules are consistency constraints, machine-readable submission is an explicit supervisory ambition. A graduate formed in formal modeling will find the domain more legible than most engineers expect — specification work with an unusually litigious client. Their gap is the tacit half: supervisory dialogue, materiality judgment, the institutional sense of what a regulator will actually accept.

Regulatory experts who learn to build. Historically the rarer direction, because building was expensive to learn mid-career. AI assistance compresses the coding half. But honesty requires the symmetric observation: regulatory knowledge is equally legible to models — the corpus is public text. Both halves compress down to tacit residues: engineering taste and failure judgment on one side, supervisory sense on the other. The remaining reason to bet on this path is error asymmetry. A domain expert’s bad code fails visibly, in review or in testing. An engineer’s bad regulatory judgment fails invisibly, until the supervisory letter arrives. Organizations that understand this will prefer to add compressible skill to the person whose mistakes are cheap to catch.

Regtech vendors as accidental finishing schools. A firm whose product is this intersection necessarily forms its employees in it. Their alumni will seed the wider market the way audit-firm alumni once seeded every finance function — the finishing-school equilibrium arriving late, from an institution nobody planned for the role.

Supervisor alumni. Regulators building suptech capability are forming both halves inside the one institution where the evidence culture is native. A small stream, but a credible one.

None of these produces a cohort quickly, which is the point of recording them now. If, in a few years, senior regulated-automation hires trace back mostly to one of these paths, that is a result. If they trace back to none of them, the parent note’s premise was wrong, and that is a result too.