01
Hypothesis
Writing copy for the portal required 10 pages at different awareness levels (1 to 5 on the Schwartz scale), varied formats (sales page, manifesto, landing, micro-copy) and a brand voice that fails itself when it drifts.
The question: does a 6-phase pipeline that routes to specialist agents produce measurably better results than one agent writing everything?
02
Method
I built COPY FORGE v3.1. 6 phases: INTAKE, INTELLIGENCE, ROUTING, DRAFT, FORGE, SHIP. 10 cognitive copywriters (Marshall, Urban, Halbert, Icaro, Bencivenga, Kennedy, Todd Brown, Ogilvy, Sugarman, Ford). Each page got a master and a specialist, chosen by awareness level and format.
Rules: Tier 0 (awareness and sophistication diagnosis) required before any draft. Voice calibrated by a reference file, with automatic fail if C1 < 0.60. QG-COPY with 5 weighted criteria. 9 pages produced in about 24 hours of sessions.
03
Result
| Page | Master | QG | C1 (voice) |
|---|---|---|---|
| Home | Marshall + Urban | 0,90 | 5/5 |
| Manifesto | Halbert + Icaro | 0,90 | 5/5 |
| Hackeando | Halbert + Todd Brown | 0,90 | 5/5 |
| Sprint | Bencivenga + Kennedy | 0,88 | 5/5 |
| Sobre | Icaro + Ford | 0,88 | 5/5 |
| Links | Ogilvy | 0,80 | 4/5 |
Average QG of Full pages: 0.89. Voice (C1): 5/5 on 100% of Full pages.
The Express pipeline (micro-copy, no agent) produced Lab, newsletter and Vault in minutes. Deciding when to use Full or Express is as critical as choosing the agents.
04
What changed
COPY FORGE became the standard production system. Agent selection now follows a routing table based on awareness × format × voice risk. The Marshall → Urban sequence (structure before texture) was validated as a repeatable pattern. A log records every decision for future audit.