Series plan: The Fifth Job
1. Name, premise, reader
Name: The Fifth Job.
Premise: In most small businesses, four of five core jobs are agent-shaped; the fifth is the business, so the decision is where agents go, not whether.
For: an engineer or Vice President (VP) deciding where agents go first, and where not.
Shared data, every post: 2,500 Starter Story businesses, each scored 0-3 per job (find, sell, deliver, support, admin) by a model reading a 300-character public description. Each post states this once, links Starter Story, and cites only name, revenue, team size and link. Gate for every foil in every post: re-read the full public story, confirm the founder still runs it, header revenue agrees with the body, team size is in the header. The five on the [[foils]] conflict list (StandOut CV, Glam Up & Sprout, Pep AI, Excel Exercises, BTNG.studio) appear only on a post's 'Deliberately left out' line, never in the body. [GATE RESOLVED 2026-10-04 17:06 ET, founder iMessage: "This is personal brand building. We are too early to consider any of this commercial use. If we develop a product or end up with phData referrals that's great, but it's not there yet." Aggregate citation with link back is OK.]
2. The spine
1. Four of five jobs are agent-shaped. The fifth is the whole business.
- Idea: the one job agents can't carry is the reason customers pay.
- Builds on: nothing; sets the frame.
- Data ([[foils]] §Handmade and artisan, §Local and in-person service, §Operations, logistics and B2B supply; [[archetypes]] §Per archetype): Frailey Woodworks, Furry Fellas, LogoClothz.
- Apply: score your five jobs 0-3; name your fifth.
- Counterpoint: the scores are a model grading itself from a blurb.
2. Admin is the first job to hand to an agent.
- Idea: start where the score is highest and mistakes cost least.
- Builds on: post 1 named the fifth job; this is the first.
- Data ([[archetypes]] §Per archetype): admin is the top job in 11 of 12 archetypes and never below 1.9 of 3 [CHECK].
- Apply: one admin workflow (invoicing, scheduling, reporting), human-checked.
- Counterpoint: admin is where the money moves; a wrong invoice costs trust fast.
3. Agents can find customers. Trust is the hard part.
- Idea: outreach and search engine optimization (SEO) are agent-shaped; being trusted is not.
- Builds on: post 1's trust, tested on finding customers (founder pushback, 2026-10-04).
- Data ([[foils]] §Marketplace and platform; [[archetypes]] §Per archetype): find averages 1.9-2.0 of 3 for software and consumer apps (saas 1.95, consumer_app 1.93), just under "does most"; every marketplace foil (Talently, Vidovo, ThirstySprout) breaks at vetting its own supply, and ThirstySprout also at client trust. [CHECK: supplier vetting is not the buyer trusting you. Either restate the idea as 'agents find; a human vouches' or pick foils where the buyer's trust is what breaks.]
- Apply: agents run outreach and content; a human vets and closes.
- Counterpoint: if every competitor's agent cold-emails, the channel dies.
4. The agency margin is payroll.
- Idea: an agency's cost is people; when agents carry production, what is left to sell is the judgment the client was paying for all along. [CHECK: observation about marketing, design and film agencies only; no restructuring advice to consultancies (phData rule, [[2026-10-02-site-spec|site-spec]]:19); keep the core belief, agents lift people up, in frame.]
- Builds on: post 3's trust, applied to services.
- Data ([[foils]] §Agency and productized service; [[archetypes]] §Per archetype; [[opportunities]] §Top 20): agencies (345 clean rows; the 12 data and AI consultancies excluded as phData-overlap, as [[opportunities]] does) score 48% carryable, deliver 1.5 of 3, $5.6k revenue per person, lowest among business-to-business archetypes; foils Carazo Creative and Layer8 against Rankings.io (169 people) and GR0 (102) from [[opportunities]]. BTNG.studio left out (conflict list).
- Apply: separate what clients pay for (judgment) from where your team's time goes (production).
- Counterpoint: clients may want the account manager, not the cheaper output. [CHECK: frame as observation; phData rule, no selling services.]
5. For anything physical, delivery is the moat.
- Idea: when the product is a thing or a visit, agents carry the business around it and not the product.
- Builds on: post 1's fifth job, at its most literal.
- Data ([[foils]] §DTC physical product brand; [[archetypes]] §Per archetype): deliver scores local 0.03, maker 0.1, ops 0.4, direct-to-consumer (DTC) brands 0.8 [CHECK]; foils Strive, No Cold Feet, Joker Greeting.
- Apply: agents on everything but the product; track hours returned to it.
- Counterpoint: revenue per person is mixed (DTC $13.0k, maker $3.0k), so "moat" doesn't always mean "money".
6. Already agent-native, and nobody noticed.
- Idea: one- and two-person businesses already score 5 of 5; the gap is know-how, not possibility.
- Builds on: posts 1-5, shown already done.
- Data ([[foils]] §Courses, info products and communities, §Content, media and audience; [[archetypes]] §Per archetype): 279 of 2,500 clean businesses score 5 of 5 with a team of one or two; foils Piano In 21 Days, Organically Addison, BBE Insider, Justin Welsh (moved from post 1). Left out: Harpagia (4 of 5, deliver 1, revenue headline-derived), Excel Exercises and Pep AI (conflict list).
- Apply: picture your team this small.
- Counterpoint: survivorship; Starter Story publishes the winners.
3. Post 1 outline
Hook: Ask an agent to run a furniture business and it will happily do the books. It will not build the table.
The frame: Small businesses do five jobs: find, sell, deliver, support, admin. I had a model score 2,500 public businesses from Starter Story on how much of each job an agent carries today. [n = 2,500 confirmed: 2,566 businesses minus 66 revenue-suspect.] Admin is the most agent-shaped job in 11 of 12 business types; deliver is the least in 9 [CHECK].
Three foils (series gate, section 1):
- Frailey Woodworks: $30k/month, 8 people, 4 of 5 jobs agent-shaped. The fifth is the furniture. [CHECK]
- Furry Fellas: $42k/month, 9 people, 4 of 5. Someone has to walk the dog. [CHECK]
- LogoClothz: $32k/month, 5 people, 4 of 5. The press, the equipment, the quality check. [CHECK]
The turn: The fifth job is not leftover work. It is why the customer pays. Sometimes it is the physical thing; sometimes it is taste, trust, or the founder's judgment (posts 3 to 5 take those one at a time). Agents carry the jobs around the business; the fifth job is the business.
What to apply:
- Score your five jobs 0 (cannot), 1 (assists), 2 (does most), 3 (runs it); name your fifth in one sentence.
- Hand agents the highest-scoring job first (post 2); protect your fifth job's hours.
The bear case: Starter Story is a self-selected sample of founder-reported wins; header numbers can be peak, stale or wrong. The scores are a model grading its own capability from a 300-character blurb; the spot-check (28 of 30 archetypes agreed, all 30 within one point) was run by the same agent that did the scoring, not by an independent reader. [BEN: re-read the 30 yourself before publishing if you want this sentence in the first person; [[archetypes]]:74.] Incumbents can adopt agents too. Confidence: moderate.
Close (to post 2): If four jobs are agent-shaped, which one do you hand over first?
Receipts: Starter Story links per foil; method note; "Deliberately left out": StandOut CV (sold April 2026; story body and header disagree).
4. One idea per post
- Series index page: premise, then six titles with one-sentence ideas; unpublished show "coming".
- "Previously" line under each byline: the prior post's idea, linked. No recap.
- Each post stands alone: restates the five jobs and data caveat in two sentences.
- One idea test: headline needs "and"? Split it; overflow ideas go to the index.
- Per [[STYLE]], every post: byline, one plain white diagram after it (CONTRACT, [[2026-10-02-site-spec|site-spec]]:28), a [BEN:] slot for a non-client anecdote, the bear case and one confidence line, a question close, receipts, the LinkedIn cut (150-180 words, acronyms re-expanded). This series is a data note, not a Snowflake build note: no invented scene, no 'what broke' H2s. [RESOLVED founder 2026-10-04: "I'll be happy with any published content." These count toward the published-pieces target.]