# Macro Context — How Much Work Is Actually Full-Automation Exposed?

*Sept 10, 2026. Strategic framing for the City Comparison project. Unlike the Bellingham phase reports (data-grounded, parsed from source documents), this memo is a WAG-level synthesis: it layers the data-grounded Bellingham findings onto economy-wide estimates from memory (BLS/BEA/Census of Governments, ~2024–25). Treat the national figures as round approximations, not citations. Lens throughout: **full automation** (remove the FTE), not copilot/assistive tools.*

---

## The one-line finding

The genuinely full-automation-exposed pool of work is **real but bounded — a ceiling around 10–15% of employment, realistically less** — and it stays bounded no matter which denominator you use. The micro (Bellingham, built role-by-role) and the macro (the whole employment pie, built top-down) converge on the same number from opposite directions.

## The number, four ways (they agree)

- **Bottom-up from Bellingham's roster:** ~7% strong full-automation candidates, ~10–15% technical ceiling, **~5–8% realistically removable** after statutory human-in-the-loop and single-incumbent constraints.
- **Top-down from the employment pie:** ~30% of jobs are desk/cognitive at all; ~half of that is *routine* enough for AI end-to-end → **~15% of employment.**
- **By dollars:** removable labor cost ≈ **$100–150B/yr** — large in isolation, ~1% of the ~$13.5T national payroll.
- **Convergence** of the bottom-up city number and the top-down economy number on ~10–15% is the most robust result of the exercise.

## Where it concentrates

Routine, codifiable, symbolic work with tolerant error bars and **no legally-required accountable human**:
- Insurance claims & underwriting back-offices
- Banking / financial operations
- Corporate administrative & finance functions
- The **junior tier** of professional services (bookkeeping, routine tax/legal doc work, junior analysts)
- Call centers
- Routine tiers of software
- A thin clerical/eligibility layer inside government and health

≈ 5–7M jobs, inside ~20% of employment, **mostly non-unionized** (differs sharply from gov/health/education).

## Why it stays bounded — three walls

1. **The employment mass is physical or interpersonal.** ~2/3 of all jobs are trades, transport, construction, manufacturing production, hands-on care, teaching, emergency response, enforcement, management, sales. A symbol-processing technology doesn't reach it.
2. **Institutional friction** — unions (~10%), occupational licensing (~21%, the bigger one), civil service, liability-driven human-in-the-loop — covers ~1/3 of workers.
3. **The accountable-human requirement** keeps a person at the decision/sign-off even where the processing automates (loan denial, claims adjudication, permit issuance, compliance principal).

## The deep structural finding: exposure ⊥ friction

Automation exposure and institutional protection sit in **opposite** places:
- The work AI is best at concentrates in **small, lightly-shielded** sectors (finance & professional back-office, information).
- The **large** sectors — government, health, education (~30% of all employment between them) — are **low-exposure AND high-friction**, shielded twice.

They are negatively correlated across the economy. That is why aggregate displacement keeps coming out below the hype: the reachable work is a minority, and it's the specific minority that is transaction-heavy but (finance's licensed front aside) not especially protected.

## The employment pie (sector sizes real; exposure/cognitive columns are estimates)

| Sector | % of jobs | Cognitive core | AI-exposed | Union+license+civil-svc friction |
|---|---|---|---|---|
| Trade, transport & utilities | ~18% | ~15% | ~10% | ~15% |
| Professional & business svcs* | ~14.5% | ~60%* | ~30%* | ~15% |
| Government | ~14.5% | ~35% | ~10% | ~55% |
| Health care & social assistance | ~14% | ~25% | ~12% | ~55% |
| Leisure & hospitality | ~11% | ~10% | ~6% | ~5% |
| Manufacturing | ~8% | ~25% | ~12% | ~12% |
| Financial activities** | ~6% | ~75%** | ~40%** | ~35% |
| Construction | ~5% | ~10% | ~5% | ~20% |
| Other services | ~4% | ~20% | ~12% | ~15% |
| Private education | ~2.5% | ~30% | ~8% | ~40% |
| Information | ~2% | ~80% | ~35% | ~10% |
| **Weighted total** | 100% | **~30%** | **~15%** | **~30%** |

\* **"Professional & Business Services" is a vestigial grab-bag** — staples ~10.5M high-cognitive professional/technical (law, accounting, engineering, software, consulting) + ~2.5M corporate management + **~9.5M administrative/support/waste (temp labor, janitorial, landscaping, security, call centers)**. ~40% of the bucket is low-wage physical labor with ~zero AI exposure. The sector-level averages hide a bimodal split; the exposed core is the *junior tier* of the professional half + call centers + corporate back-office.

\*\* **Financial activities: regulated ≠ protected.** Its regulation is mostly *process/institutional* (SOX, Dodd-Frank, BSA/AML, fair-lending, GLBA, state insurance codes) — codified, auditable rules, which is **automation-friendly substrate, not a brake** (hence RegTech). The real friction is *individual licensing* (brokers, agents, mortgage originators, CPAs) + liability requiring an accountable human — both at the front/decision layer, not the processing layer. So the unlicensed back office is **both heavily regulated and highly automatable** (~50–55% exposed); the licensed/relational front (advisors, real-estate agents) is protected (~15–20%). Insurance is the softest underbelly: a giant claims-and-underwriting processing operation.

## Methodological punchline (the case for this whole project)

You cannot see any of this from the schemas. BLS bundles janitors with law partners; "Financial Activities" hides an automatable back-office behind a licensed front; the city budget renders its own position tables as scrambled character-runs. Inherited categories are accounting conveniences that **actively mislead** on automation. The only way to a real answer is the **task / duty-statement level** — which is exactly why Bellingham was built role-by-role. The city was the right place to *learn the method* because it's the most legible org in existence — even though it turned out to be near the *least* exposed (mostly water pipes and fire trucks).

## Caveats (so "limited" isn't over-read)

- **Copilot footprint is far larger than full automation.** Assistive AI touches the whole ~30% cognitive slice, raising output-per-worker rather than cutting headcount. Full automation understates AI's real footprint considerably.
- **WAGs stacked on WAGs.** The soft layer is the per-sector exposure estimates. Running real O*NET task data — or Bellingham's 352 duty statements — through a proper scorer would firm this a lot.
- **Two frontier-movers:** robotics is the wildcard that could eventually reach the physical 2/3 (different, slower timeline); and the "accountable-human" wall is partly a *policy* variable, not a fixed technical one — law could move it either way.

## Strategic implication for the project

If the goal shifts from "where is government most *legible*" to "where is the automation *exposure*," the high-yield targets are **transaction-heavy administrative bodies** — benefits and eligibility agencies, insurance operations, financial back-offices — **not general-purpose cities**. Bellingham (and Everett, Spokane Valley) remain the right specimens for *validating the method*; they are not where the impact is. The method transfers; the target selection for impact should tilt toward the administrative/eligibility tail.