Agent & Instrument Diagram

How do agents orchestrate a skill into an operating business?

Agents

Agent	Type	Code	Authority	Role	Phase
Business Idea Generator	Orchestrator	`OA[L2]`	Decide	Coordinates full flow from input to operating venture	All
Domain Researcher	AI Worker	`DA[L1]`	Execute	Market analysis, competitor research, friction identification	1
Eval Runner	Quality Gate	`DA[L1]`	Execute	LLM-as-Judge validates AI-generated components	1-2
Sprout Generator	Infrastructure	`DA[L1]`	Execute	Creates technical substrate when validation gate passes	1
Operations Seeder	Automation	`DA[L1]`	Execute	Populates Work Charts, sales pipeline, marketing cadence	2
Agent Workforce	Delegation	`DA[L1]`	Execute	AI agents assigned to WorkChart stages for ongoing operations	3
Human (Founder)	Override	`HA[L3]`	Override	Can halt or redirect any gate, any agent	All

Instruments

Instrument	Code	Function	Measures	Action
VFL Scorer	`VFL-G01`	Gate	Idea alignment with VVFL criteria	`[P]` proceed; `[F]` stop or pivot
Optimal Stopping	`OPT-G02`	Gate	Research signal sufficiency (37% rule)	`[P]` create venture; `[F]` keep researching
Explore-Exploit	`EXP-S01`	Selector	Channel/ICP/positioning effectiveness	Exploit best; Explore new when stale
SPCL Scoring	`SPCL-M01`	Monitor	Content authority potential per topic	Highest score → publish first
Compound Rate Tracker	`CPD-M02`	Monitor	Reusable components added back to platform	Growing → healthy; Flat → extractive
Sales Forecasting	`SF-R01`	Recorder	Deal probability for domain	Logs projections for feedback

Flow

PERSON WITH A SKILL
       │
       ▼
┌──────────────────────────┐
│  OA[L2] ORCHESTRATOR     │
│  Business Idea Generator │
└──────────┬───────────────┘
           │
    ┌──────┴──────┐
    ▼              ▼
DA[L1]          DA[L1]
DOMAIN          EVAL RUNNER
RESEARCHER      (quality gate on
(AI market       every AI output)
 analysis)           │
    │                │
    └──────┬─────────┘
           ▼
   ┌───────────────┐
   │  VFL-G01      │ ←── Gate Instrument
   │  Is this idea │
   │  virtuous?    │
   └───────┬───────┘
           │
    [P]    │   [F] → Stop / Pivot
           ▼
   ┌───────────────┐
   │  OPT-G02      │ ←── Gate Instrument
   │  Enough       │
   │  signal?      │
   └───────┬───────┘
           │
    [P]    │   [F] → Keep researching
           ▼
   ┌───────────────┐
   │  DA[L1]       │ ←── Infrastructure Agent
   │  SPROUT       │
   │  GENERATOR    │
   └───────┬───────┘
           │
           ▼
   ┌───────────────┐
   │  DA[L1]       │ ←── Automation Agent
   │  OPERATIONS   │
   │  SEEDER       │     Instruments:
   │  Work Charts  │     - SPCL-M01 (Monitor)
   │  Sales        │     - EXP-S01 (Selector)
   │  Marketing    │     - SF-R01 (Recorder)
   └───────┬───────┘
           │
           ▼
   ┌───────────────┐
   │  DA[L1]       │ ←── Delegation Agent
   │  AGENT        │
   │  WORKFORCE    │     Instruments:
   │  AI handles   │     - LR-C01 (Controller)
   │  admin,       │     - CPD-M02 (Monitor)
   │  human        │
   │  handles      │
   │  craft        │
   └───────────────┘

   HA[L3] ⊗ VFL-G01    Human can override the VFL gate
   HA[L3] ⊗ OPT-G02    Human can override Optimal Stopping

Feedback Loops

Loop	Type	From	To	Signal	Frequency
Algorithm Learning	`↺`	WorkChart receipts	VFL-G01 + Eval Runner	Did the scored idea produce real value?	Per receipt
Cross-Venture	`⟳`	Venture A receipts	Venture B templates	Which templates work for similar domains?	Per venture
Agent Tuning	`↺`	Agent performance data	LR-C01	Which agent configs produce better outcomes?	`[T:weekly]`
Compound Rate	`↺↺`	Reusable components	Platform template library	Is this venture contributing back?	`[T:monthly]`
Founder Feedback	`↺`	Human judgment	ICP + Value Prop refinement	Did the generated business match reality?	Per milestone

The Meta-Algorithm

The venture creation flow IS the Universal Algorithm:

INTENT (person has a skill + sees a gap)
  → ROUTE (OA[L2] scores idea via VFL-G01, OPT-G02)
  → INFRASTRUCTURE (DA[L1] Sprout Generator builds substrate)
  → SETTLE (venture launches, first Work Charts execute)
  → FEEDBACK (outcomes feed VFL-G01 via ↺, improving next venture via ⟳)

Every venture that runs on the mycelium makes the next venture better. The algorithms learn from aggregate WorkChart receipts across all ventures.

Key Finding

The acceptance test for the entire pipeline is the first ⟳ learning loop completing: a real founder inputs a real skill → the generator produces a real venture → the venture executes real Work Charts → the receipts improve the generator for the next founder. Until that loop closes, every component is an assumption. VFL-G01 and OPT-G02 are the two critical instruments — they decide what gets built and when. If those two gates work, the rest is plumbing.

Context

Capability Map — Previous: what can we actually do?
Pictures Index — Hub: all five maps
A&ID Template — Nomenclature reference (v2.0)

Agents​

Instruments​

Flow​

Feedback Loops​

The Meta-Algorithm​

Key Finding​

Context​