BusinessRnD · draft v2 · built 2026-10-21 06:15 UTC

The machine, end to end

Cards have Go deeper, Maybe and Kill buttons. Press Send feedback to save. Your marks from all tabs collect at the end of each page.

Design

One harness pattern runs every service: detect, sample, email, take payment, deliver, check, upsell. You only check quality and handle money.Your time (est.): ~35–55 h for the first 50 sales, then ~15–25 h/month at ~50 sales · tools ~$300–600/month

The loop

flowchart TD
  A["Detector scores public data"] --> B["Proof builder makes the free sample"]
  B --> C["Outreach agent sends the email"]
  C --> D{"Reply?"}
  D -->|"yes"| E["AI answers, sends pay link"]
  D -->|"no"| X["One follow-up, then stop"]
  E --> F["Intake form: access and preferences"]
  F --> G["Producer pipeline"]
  G --> H["Automatic checks"]
  H --> I["Your 10–15 min check"]
  I --> J["Delivery with before and after report"]
  J --> K["Monthly plan offer"]

Who does what

AIEverything repeatableFinding businesses, samples, emails, replies, intake, production, automatic checks, reports, follow-ups.

YouQuality and money10–15 min check per job, refunds and escalations, weekly withdrawal.

ClientPay and give accessClicks the pay link, fills the form, approves the delivery.

Quality gates (before and during selling)

Gate	Threshold	Why
Before selling a service	≥70% of 30 real samples "ship as-is"	Agents score ~21% on broad real projects; only narrow work passes
Every job	Automatic checks pass, then your 10–15 min look	Catches theme breaks, artefacts, invented facts
Live	First-pass acceptance ≥60%, refunds ≤15%	Below this, stop selling and fix

Where the AI compute runs

Step	Runs on	Who pays
Building and improving harnesses, quality checks	Your Claude Code subscription	Already paid
Pilot production (first tens of jobs)	Your subscription, within its limits	Already paid
Production at volume	Per-use non-Anthropic models (D16)	~$4–15 per job, from revenue
Video generation (S3 only)	Veo-class models	~$50–260 per pack

Why not the subscription at volume? Limits assume individual use, weekly caps are unpublished and keep changing, and Anthropic says services should use API keys. Your rule forbids API credits, so volume moves to other providers. Max's bundled API credits stay unused unless you decide otherwise.

The money at steady state (estimate)

Assumption (from the round-3 notes)	Result
~50 sales a month, mixing $249 and $499
~20% of buyers on a $149 monthly plan
After tax (~~6%), payment fees (~~4%), compute and tools	~$10k profit, ~15–25 owner hours

At a ~$600 average ticket (e.g. the accessibility pack), the same profit needs only about 20 jobs a month.

What breaks it

Samples don't convert: the core thesis, measured in the first 1,000–3,000 contacts.
Deliverability: burned domains. Mitigated by caps, warm-up and opt-outs.
Chargebacks on Payoneer links. One in the first 50 orders means wire-only.
Scope creep: fixed deliverables, one revision, "choose 6 of 10" for ads.
Legal claims: never "compliant" for accessibility, no invented testimonials, no guaranteed rankings.

Sources

Round-3 notes: economics, proof and risks — research_notes/Round 3 agent services/economics_proof_risks.md
Round-3 notes: productized proof and sales — research_notes/Round 3 agent services/productized_proof_sales.md
Round-3 notes: job types and harness designs — research_notes/Round 3 agent services/job_types.md

Your feedback

Overall on this tab: what to change, what to dig into, what to drop

What you marked so far (all tabs)

Nothing yet.

Feedback is stored on this device and sent to the report's private storage. If sending fails, copy it and paste it into the chat.