The decomposition is borrowable from APQC, the exposure score is house-original, and the unit of analysis is the part no published framework can validate
The question
Verbatim from the Notion Research Backlog (page_id 3f9f7d4936d181a9b62acda3272e90bb): "Is 'agent-shaped job' (agentify-3000's core scoring taxonomy — find/judge/do/deliver/admin, scored 1-3 per job per archetype) grounded in any published task-automation-exposure framework (O*NET task-level automation exposure, Autor/Frey-Osborne-style labor literature), or is it a house-original taxonomy with no outside validation?"
The question misstates the taxonomy, and the correction matters. The live scheme in data/scripts/phase2/scheme.md is find / sell / deliver / support / admin, scored 0-3 (not 1-3), per business (2,566 of them), with archetype-level figures being means of business scores. There is no judge axis and no do axis. Everything below scores the real five. The corrected statement is also the first finding: if the queue note drifted off the rubric in two weeks, a published post will too.
What we already know (from the vault)
- The rubric, verbatim. Five core jobs scored 0-3 "for how much an AI agent can carry TODAY (October 2026) in THIS business, given current agent capability (web and software actions, writing, voice and chat, data work, drafting and scheduling; no robots, no legal standing, no physical presence)." Anchors:
0= cannot carry (physical hands, licensed signature, in-person trust, taste the market pays for) ·1= assists (drafts, schedules, summarizes; a human does the core act) ·2= does most, human reviews/approves/handles exceptions ·3= runs it end-to-end with light oversight. Jobs:find(demand gen, SEO, outbound, ads, listings) ·sell(qualify, demo, quote, close, checkout, onboarding, renewals) ·deliver(the core production or fulfillment work) ·support(tickets, chat, returns, retention) ·admin(bookkeeping, invoicing, hiring and scheduling, compliance, reporting, vendors). Two derived fields ride on top:agent_share(0-100, hours-weighted, explicitly "not a plain mean of the five scores") andmoat(one phrase naming what a human still has to bring). Source: [[archetypes]], [[IMPLEMENTATION-NOTES]]. - The headline it produces. 12 archetypes, overall median agent-carryable share 45%;
adminis the top job in 11 of 12 archetypes and never below 1.9;deliveris the floor for anything physical (local 0.03, maker 0.1, ops 0.4, coaching 0.7, dtc_brand 0.8). See [[archetypes]]. - Validation to date is author-agreement, not inter-rater reliability. [[IMPLEMENTATION-NOTES]] records a 30-business hand check: archetype 28/30 (93%), "job scores: 30/30 within one point of my own read on every job,"
agent_share30/30 within 15 points. The comparison rater is the person who wrote the rubric, and no chance-corrected statistic is reported. - It is already load-bearing on a public surface. The Ship 30 week-1 draft states the method and the
find= 1.9 mean in a published-facing post and still carries an inline[BEN: confirm you ran this scoring]; the series plan commits every post to restating the method once. See [[agentify-3000-series]], [[2026-10-04-starter-story-pat-walls-assessment]]. - The vault has already run this exact play and lost it once. [[industry-process-reference-models]] is the record of RDCO systematically testing "borrow an established taxonomy vs build our own" against APQC, BIZBOK, SCOR and eTOM for the Organizational Platform — the recurring negative was that general frameworks decompose functions reliably and say nothing about the layer above. That is precisely the shape of the answer here, inverted.
What the web says
O*NET publishes the task taxonomy and no exposure score
Fetched onetcenter.org/content.html directly. Work-activity content sits under Occupational Requirements → Work Activities ("41 variables / 74,702 ratings / 2,085 detailed work activities / 24,087 DWA-task linkages … This hierarchy includes generalized, intermediate, and detailed work activities"). Tasks is a separate domain — Occupation-Specific Information — with 18,838 statements. The four top-level groupings beneath Work Activities are, verbatim: Information Input ("Where and how are the information and data gained…"), Mental Processes ("What processing, planning, problem-solving, decision-making, and innovating activities are performed…"), Work Output ("What physical activities are performed, what equipment and vehicles are operated/controlled…"), Interacting With Others ("What interactions with other persons or supervisory activities occur…").
Two things follow. First, O*NET itself publishes no automation- or AI-exposure score per task — confirmed against the page text; the one near-miss is Work Context descriptor 4.C.3.b.2 "Degree of Automation — How automated is the job?", a survey-collected incumbent rating of the job as it is today, not a forward-looking per-task estimate. O*NET is the substrate everyone else scores. Second, O*NET's four groupings are cognitive-stage categories, not business-function categories — and they are an uncanny match for the axis set the queue note hallucinated (find→Information Input, judge→Mental Processes, do→Work Output, deliver→Interacting With Others). Those four have outside authority. The five we actually ship do not, on this axis.
Every exposure framework scores tasks and rolls up to occupations
- Autor, Levy & Murnane, Quarterly Journal of Economics 118(4): 1279-1333, Nov 2003 (NBER WP 8337). Verified claim from the NBER abstract: computer capital "(1) substitutes for a limited and well-defined set of human activities, those involving routine (repetitive) cognitive and manual tasks; and (2) complements activities involving non-routine problem solving and interactive tasks." The routine × cognitive/manual grid is confirmed; the specific five category labels commonly attributed to ALM could not be verified (OUP returns 403 to both WebFetch and curl) and are not asserted here.
- Frey & Osborne, "The Future of Employment: How Susceptible Are Jobs to Computerisation?", working paper dated 17 September 2013 (Oxford Martin School). 702 detailed occupations. Method verified from the PDF: "we subjectively hand-labelled 70 occupations, assigning 1 if automatable, and 0 if not," then a Gaussian process classifier (exponentiated-quadratic covariance selected, AUC 0.894) produced a 0-1 probability for all 702. Three engineering bottlenecks, verbatim: Perception and Manipulation, Creative Intelligence, Social Intelligence — proxied by nine named O*NET variables rated on the level scale (Finger Dexterity, Manual Dexterity, Cramped Work Space; Originality, Fine Arts; Social Perceptiveness, Negotiation, Persuasion, Assisting and Caring for Others). Headline: "about 47 percent of total US employment is at risk." Journal version Technological Forecasting and Social Change 114: 254-280 (2017) is medium confidence — inferred from a RePEc URL slug, not fetched page text.
- Brynjolfsson, Mitchell & Rock, "What Can Machines Learn and What Does It Mean for Occupations and the Economy?", AEA Papers and Proceedings 2018, 108: 43-47. The construct is "suitability for machine learning" (SML). Verbatim on the instrument: "The rubric we apply has 23 distinct statements to be evaluated on a 5-point scale varying from 'strongly disagree' to 'strongly agree.'" Scored unit is the DWA, not the task: "We use the O*NET content model for 964 occupations … joined to 18,156 specific tasks … which are further mapped to 2,069 direct work activities (DWAs)." Raters were CrowdFlower crowdworkers. (The same published PDF reports 966 occupations / 19,612 tasks in Table 1 — an unresolved internal discrepancy.) Their distribution finding is the load-bearing one for us: "most occupations in most industries have at least some tasks that are SML; few if any occupations have all tasks that are SML," and the field should move "away from the common focus on full automation of many jobs and pervasive occupational replacement toward the redesign of jobs and reengineering of business processes."
- Eloundou, Manning, Mishkin & Rock, "GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models", arXiv:2303.10130 (v1 17 Mar 2023; v5 21 Aug 2023; stamped WORKING PAPER, no journal venue on the arXiv record). This is the closest formal relative of our rubric. Four ordinal levels with an operational threshold:
E0no exposure (no or minimal time reduction at equivalent quality, or quality decreases) ·E1direct exposure — "can decrease the time required to complete the DWA or task by at least half (50%)" ·E2LLM+ Exposed — needs additional software built on the LLM to hit the same 50% ·E3image capabilities. Critical: "we combine E2 and E3 for all analyses," so three tiers are reported, not four. Applied to O*NET 27.2 (1,016 occupations, 19,265 tasks, 2,087 DWAs) by both human annotators and GPT-4; headline numbers use "GPT-4, Rubric 1." Human-vs-GPT-4 agreement on E1 is 80.8% raw but kappa 0.223 — weak once chance-corrected. - Agent-era instruments (2025-2026). Anthropic Economic Index maps conversations to O*NET tasks via Claude-as-classifier and splits interactions into automation (Directive, Feedback Loops) vs augmentation (Learning, Task Iteration, Validation). The January-2026 report (Nov 13-20 2025 data, consumer Claude.ai) gives augmented 52% / automated 45%; a June 2026 report exists and reportedly redefines "automation share," so the current split is unverified here — do not cite 52/45 as today's number. OpenAI GDPval (arXiv:2510.04374, Oct 2025) scores task deliverables against expert-produced deliverables across "BLS Work Activities for 44 occupations across the top 9 sectors"; grading mechanics unverified. Remote Labor Index (arXiv:2510.26787, Oct 2025) scores real freelance-style projects end-to-end, best agent automation rate 2.5%.
Nobody scores a business
A deliberate negative search returned nothing that assigns an automation-exposure score to a firm or business entity: arXiv metadata queries for firm-level AI/automation exposure returned zero entries, and Anthropic states its index "does not map conversations to specific businesses or firms" (firm behaviour is inferred from first-party API traffic as a proxy, where "three-quarters of interactions classified as automation"). Every verified instrument indexes task → occupation (SML, Frey-Osborne, ALM, Anthropic) or task/project → deliverable (GDPval, RLI). Caveat: this negative rests on arXiv metadata search plus one fetched page; consultancy and vendor indices were not swept, so it is real but not conclusive.
Convergences and contradictions
- The decomposition converges with APQC, not with the labor literature. [[2026-08-17-apqc-pcf-function-motion-mapping]] read APQC's Process Classification Framework v7.4 (August 2024) in full from the primary PDF, and its 13 categories split exactly as our five axes do. PCF's own operating/management-and-support line is the same line our
adminaxis draws. - The scoring diverges from every published rubric on unit, not on scale. 0-3 four-level ordinal sits comfortably inside the published range (Frey-Osborne binary→continuous, Eloundou four-collapsed-to-three, SML 5-point Likert ×23 items). What no published rubric does is score a business function inside one firm. That is the house-original move.
- Our headline finding is more grounded than our taxonomy. "Four of five jobs are agent-shaped, the fifth is the whole business" is the firm-level restatement of Brynjolfsson/Mitchell/Rock's verified claim that most occupations have some high-SML tasks and almost none have all of them. We can cite that. We cannot cite anyone for the five axes.
Mapping verdict, axis by axis
| Axis | Closest named published construct | Verdict |
|---|---|---|
| find | APQC PCF 3.0 Market and Sell Products and Services, level-2 groups 3.1 Understand markets, customers and capabilities · 3.2 Develop marketing strategy · 3.3 Develop and manage marketing plans | Named construct (process framework). No labor-literature analogue. |
| sell | APQC PCF 3.4 Develop sales strategy · 3.5 Develop and manage sales plans | Named construct. Our find/sell split is a sub-split inside one PCF category — defensible, and PCF itself draws the same seam at level 2. |
| deliver | APQC PCF 4.0 Manage Supply Chain for Physical Products + 5.0 Deliver Services | Defensible recombination: two PCF categories collapsed into one axis. Justified for small businesses that do one or the other, not both. |
| support | APQC PCF 6.0 Manage Customer Service | Named construct, effectively 1:1. The strongest single mapping we have. |
| admin | APQC PCF categories 7.0-13.0, i.e. APQC's entire explicitly-labelled "Management and support services" block (human capital, IT, financial resources, assets, risk/compliance, external relationships, business capabilities) | Defensible recombination, and the weakest axis: it collapses seven PCF categories into one score. Its 2.0-2.3 means across every archetype are near-certainly an artifact of that aggregation, not a finding. |
| (missing) | APQC PCF 1.0 Develop Vision and Strategy and 2.0 Develop and Manage Products and Services | No axis exists. And the rubric knows it: scheme.md's own agent_share instruction says "a SaaS's are build, support, find" — naming a sixth job the taxonomy does not contain. This is an internal inconsistency, not a philosophical gap. |
| moat field | Frey & Osborne's three engineering bottlenecks: Perception and Manipulation · Creative Intelligence · Social Intelligence | Near-exact match, and the most citable thing in the whole scheme. Our 0 anchor — "physical hands, licensed signature, in-person trust, taste the market pays for" — is Frey-Osborne's three bottlenecks plus licensure/legal standing, which F&O does not include. That fourth item is a genuine extension, not a restatement. |
| 0-3 scale | Eloundou et al.'s E0/E1/E2(+E3) ordinal exposure rubric | Granularity defensible; anchors are not operational. Eloundou pins each tier to a measurable threshold (≥50% time reduction at equivalent quality, where "equivalent quality means a third party … would not notice or care about LLM assistance"). Ours are share-of-act descriptions with no threshold, so two raters can disagree without either being wrong. |
| unit of analysis (business, not occupation) | none | No outside analogue, qualified. Verified absence across the academic preprint surface; consultancy indices unswept. |
Overall verdict: partially grounded. The decomposition is a defensible recombination of a named, current, primary-source-verified process framework (APQC PCF v7.4) — not a house invention, and we should stop treating it as one. The exposure scoring is house-original: no published framework scores business functions, the anchors have no operational threshold, and the 30-business check is author-agreement rather than independent inter-rater reliability. The moat field and the 0 anchor are the parts with the cleanest outside authority, and they are currently the parts we never cite.
Synthesis for RDCO
Can the scoring engine cite outside authority today? Partly, and not where it matters. Right now a data-literate reader of the Fifth Job series gets five axes, a 0-3 scale, and a 2,566-row n, with no named framework behind any of it and no reliability statistic. That reader's first move is the one Eloundou et al. pre-empted on themselves and we have not: who else scored these, and did they agree? Our answer is "the author checked 30 and liked them." Eloundou reports 80.8% raw agreement on direct exposure and a kappa of 0.223 on the identical problem shape — model-labelled exposure scoring against human labels, at scale, by the people who built the model. If the published state of the art admits weak chance-corrected agreement in print, a founder-run study claiming 30/30-within-one-point without a kappa is not reading as more rigorous; it is reading as unaudited. That is the credibility exposure, and it lands squarely on the "agents in production" positioning the phData sales motion is supposed to borrow from ([[2026-10-08-verification-ratio-senior-signoff-throughput]]).
The cheap fix is three citations and one number, and it is almost entirely rewriting, not rescoring. (1) Anchor the five axes to APQC PCF v7.4 explicitly — one sentence per post: "the five jobs are APQC's Process Classification Framework categories 3-6 plus its management-and-support block, collapsed for businesses under 20 people." That converts the taxonomy from invented to derived from a 1992-vintage open standard built with 80+ organizations, which is a materially different sentence to defend. The primary-source read is already in the vault ([[2026-08-17-apqc-pcf-function-motion-mapping]]), so this costs nothing. (2) Restate the 0-3 anchors as Eloundou-style time-reduction bands with a quality clause, and cite arXiv:2303.10130 for the threshold convention. (3) Cite Brynjolfsson/Mitchell/Rock for the finding, not the method — their "few if any occupations have all tasks that are SML … toward the redesign of jobs and reengineering of business processes" is our thesis one level up, and it is the single strongest external validation available to the series. And: run a held-out sample of 50-100 businesses past an independent rater and publish a weighted kappa. One number, generated once, that nobody else in this genre publishes.
The unit of analysis is the asset, and we have been apologizing for the wrong thing. Every verified instrument — ALM, Frey-Osborne, SML, Eloundou, Anthropic Economic Index, GDPval, RLI — indexes tasks and rolls up to occupations, projects, or deliverables. None scores a business. That is not a gap in our rigor; it is the whitespace. The economics literature answers "which jobs are exposed," which is a policy question. agentify-3000 answers "which businesses can be run by one person and an agent fleet," which is an operator question, and it is the one a phData prospect and a Sanity Check reader both actually have. The same pattern RDCO already banked in [[industry-process-reference-models]] applies: borrow the decomposition, own the layer above it. Here the layer above is the firm-level aggregation (agent_share, hours-weighted, deliberately not a mean) and the moat phrase. Those are the original contributions and they should be the headline, with the five axes positioned as borrowed scaffolding rather than as the insight.
Two defects to fix before anything else publishes. First, the missing axis: scheme.md names "build" in its own agent_share instruction while the taxonomy has no build/product-development job, which maps to PCF 2.0 and matters enormously for the saas and consumer_app archetypes that currently score highest. Any post claiming a SaaS is 58% agent-carryable is claiming it about a business whose largest hour sink was never scored. Second, admin collapses seven PCF categories, which almost certainly manufactures its own result — "admin is the universally agent-shaped job, top in 11 of 12 archetypes" is exactly what you would expect from averaging seven heterogeneous functions, and it is currently finding #1 in [[archetypes]]. Neither defect requires rerunning 2,566 businesses to state; both require stating before the series leans on them.
Why this is in the vault
agentify-3000's five-axis score is the engine behind the Fifth Job Ship 30 series ([[agentify-3000-series]]) and the opportunity ranking in opportunities.md, and the series is one of the few surfaces actively building the "agents in production" credibility the founder named as the phData sales lever. This brief is the grounding check that decides whether each post can cite APQC + Eloundou + Brynjolfsson or has to say "a taxonomy I made up," and it flags two scoring defects (no build axis, admin over-aggregation) that the series' two headline findings currently rest on.
Open follow-ups
- Does an independent rater reproduce the 0-3 job scores on a held-out sample at acceptable weighted kappa? Unanswered and unanswerable from literature — it is a measurement we have to run on ourselves.
- Do any non-academic firm-level automation-exposure indices exist (McKinsey/BCG/Gartner-style, or vendor "AI readiness" scores)? The clean negative above covers the preprint surface only, and the whitespace claim should not go public until that sweep is done.
- What are Brynjolfsson/Mitchell/Rock's 23 SML items verbatim? They are in the AEA supplementary materials / OpenICPSR replication package 114436 / the Stanford Digital Economy Lab "WorldSML" page. If several map onto our 0-3 anchors, the anchors get a second citation for free.
- What does the June 2026 Anthropic Economic Index report say, and how did it redefine "automation share"? The current automation/augmentation split is unverified here, and it is the one agent-era number the series would most want to quote.
- Autor/Levy/Murnane's exact task-category labels remain unverified (OUP 403). Worth closing only if a post wants to invoke the routine/non-routine frame by name.
Related
- [[archetypes]] — the per-archetype table and the five-job means this brief audits
- [[IMPLEMENTATION-NOTES]] — the verbatim rubric, the classification run, and the 30-business hand check
- [[agentify-3000-series]] — the publishing surface that inherits this credibility exposure
- [[2026-08-17-apqc-pcf-function-motion-mapping]] — primary-source read of APQC PCF v7.4, the framework the five axes should cite
- [[industry-process-reference-models]] — the vault's prior "borrow vs build a taxonomy" round, same shape, opposite conclusion
- [[2026-10-04-starter-story-pat-walls-assessment]] — the dataset's provenance and ToS constraints
- [[2026-07-27-psf-pyramid-economics-post-agent-analyst-tier]] — the vault's other firm-level agent-absorption analysis
- [[2026-10-09-occupational-exit-rate-distribution-cut-point]] — the occupation-level side of the AI-labor cluster this brief connects to
- [[2026-10-08-verification-ratio-senior-signoff-throughput]] — the positioning bet that this credibility question feeds
Sources
Vault
01-projects/agentify-3000/data/scripts/phase2/scheme.md(the rubric as run)01-projects/agentify-3000/archetypes.md→ [[archetypes]]01-projects/agentify-3000/IMPLEMENTATION-NOTES.md→ [[IMPLEMENTATION-NOTES]]01-projects/personal-brand/series/agentify-3000-series.md→ [[agentify-3000-series]]01-projects/personal-brand/ship30/2026-10-07-week1-drafts.md06-reference/research/2026-08-17-apqc-pcf-function-motion-mapping.md→ [[2026-08-17-apqc-pcf-function-motion-mapping]]06-reference/concepts/industry-process-reference-models.md→ [[industry-process-reference-models]]06-reference/2026-10-04-starter-story-pat-walls-assessment.md→ [[2026-10-04-starter-story-pat-walls-assessment]]06-reference/research/2026-07-27-psf-pyramid-economics-post-agent-analyst-tier.md→ [[2026-07-27-psf-pyramid-economics-post-agent-analyst-tier]]06-reference/research/2026-10-09-occupational-exit-rate-distribution-cut-point.md→ [[2026-10-09-occupational-exit-rate-distribution-cut-point]]06-reference/research/2026-10-08-verification-ratio-senior-signoff-throughput.md→ [[2026-10-08-verification-ratio-senior-signoff-throughput]]
Web (fetched and read; quoted text is from these documents)
- O*NET Content Model — https://www.onetcenter.org/content.html
- Frey & Osborne (2013), "The Future of Employment" — https://oms-www.files.svdcdn.com/production/downloads/academic/The_Future_of_Employment.pdf
- Brynjolfsson, Mitchell & Rock (2018), AEA P&P 108: 43-47 — https://www.cs.cmu.edu/~tom/pubs/AEA2018-WhatCanMachinesLearn.pdf
- Eloundou, Manning, Mishkin & Rock, "GPTs are GPTs", arXiv:2303.10130v5 — https://arxiv.org/abs/2303.10130
- Autor, Levy & Murnane, NBER WP 8337 (published QJE 118(4): 1279-1333, 2003) — https://www.nber.org/papers/w8337
- Anthropic Economic Index, January 2026 report — https://www.anthropic.com/research/anthropic-economic-index-january-2026-report
- Patwardhan et al., "GDPval", arXiv:2510.04374 — https://arxiv.org/abs/2510.04374
- Mazeika et al., "Remote Labor Index", arXiv:2510.26787 — https://arxiv.org/abs/2510.26787
Could not verify (flagged, not guessed)
- Autor/Levy/Murnane's exact task-category label set and count — academic.oup.com returns 403 to WebFetch and curl; the routine × cognitive/manual structure is verified from the NBER abstract, the five commonly-cited labels are not.
- Frey & Osborne journal placement (TFSC 114: 254-280, 2017) — inferred from a RePEc URL slug, not fetched page text. Medium confidence.
- Item count of the original Brynjolfsson & Mitchell (2017) Science 358(6370): 1530-34 rubric — paywalled. The 2018 paper calls its 23-item version "slightly extended," so 23 may not apply upstream.
- The 23 SML items themselves — in AEA supplementary materials / OpenICPSR 114436 / Stanford DEL "WorldSML". Not fetched.
- O*NET's descriptor rating-scale anchors and numeric ranges — the string "scale" does not occur on content.html; they live in the questionnaires/data dictionary, not fetched.
- GDPval's grading mechanics (win/tie/loss rates, grader rubric) — openai.com returns 403; paper body not read.
- Remote Labor Index methodology beyond the abstract (project count, sector list, adjudication of "automation rate").
- The current Anthropic automation/augmentation split — a June 2026 report exists and reportedly redefines "automation share"; the 52%/45% figure here is Nov-2025 data from the January-2026 report and must not be presented as current.
- Whether any non-academic (consultancy/vendor) firm-level automation-exposure index exists — not searched; the "nobody scores a business" negative covers the academic preprint surface only.
Research budget note: QMD 2 queries + direct grep/read of 8 vault docs; web 3 WebSearch + 4 WebFetch across two verification subagents (one fetch over the template's 3-source cap, spent to verify the SML scale rather than guess it). Paywalled sources were skipped and flagged, never inferred.