What is FCC?

This tutorial introduces the FCC (Find-Create-Critique) Agent Team Framework: what problem it solves, how it approaches the solution, and what the three phases look like in practice.

The Problem

Documentation in software projects suffers from predictable failures:

1. No structured research phase. Writers jump straight into drafting without systematically gathering requirements, existing knowledge, or stakeholder input. The result is documentation built on incomplete understanding.

2. No separation of creation concerns. A single person or team handles everything from API specifications to user guides to runbooks, with no specialization. Architectural blueprints get the same treatment as onboarding tutorials.

3. No systematic critique. Review is ad hoc -- a quick read-through before publication. There is no formal validation against quality standards, no governance audit, no fact-checking of AI-generated content.

4. No feedback loops. When a runbook reveals a gap in the architecture specification, there is no structured path to feed that finding back into research and design.

The result: documentation that is inconsistent, incomplete, unvalidated, and quickly outdated.

The Solution: Find-Create-Critique

FCC addresses these failures by separating the documentation lifecycle into three distinct phases, each staffed by specialized AI agent personas:

graph LR
    F[Find] -->|Research outputs| C[Create]
    C -->|Draft artifacts| CR[Critique]
    CR -->|Feedback| F
    CR -->|Feedback| C

Find Phase

The Find phase is dedicated to research, inventory, and knowledge gathering. No content is created in this phase -- the goal is to build a comprehensive knowledge base that downstream personas will consume.

What happens:

• The Research Crafter (RC) gathers and organizes all relevant information: source code, architecture docs, stakeholder feedback, operational procedures
• The Catalog Indexer Architect (CIA) indexes all documentation assets into a searchable catalog
• The Semantic Taxonomy Engineer (STE) establishes consistent terminology using triplet logic (subject-predicate-object)
• The Research Inventory Crafter (RIC) automates capability matrix creation in machine-parseable formats

Outputs: Capability matrices, research inventories, annotated references, traceability matrices, index schemas, taxonomy schemas.

Create Phase

The Create phase transforms research outputs into polished documentation artifacts. Each persona specializes in a specific type of output.

What happens:

• The Blueprint Crafter (BC) translates research into design documents, API specifications, data models, and workflow definitions
• The Runbook Crafter (RB) produces step-by-step operational procedures with automation scripts
• The User Guide Crafter (UG) creates accessible, actionable user documentation
• The UI Mockup Crafter (UMC) generates visual prototypes and design specifications
• The Privacy Taxonomy Engineer (PTE) classifies data by sensitivity level
• The Executive Communicator (EC) translates technical content for leadership audiences

Outputs: Design documents, API specifications, runbooks, user guides, UI mockups, executive summaries.

Critique Phase

The Critique phase validates, reviews, and governs all artifacts before publication. This is where quality is enforced.

What happens:

• The Documentation Evangelist (DE) enforces style guide compliance and documentation standards
• The Blueprint Validator (BV) verifies blueprint completeness against quality gate thresholds
• The Anti-fact Mitigation Specialist (AMS) validates AI-generated content against authoritative sources using confidence scoring
• The Governance Compliance Auditor (GCA) audits all artifacts against governance frameworks

Outputs: Review reports, validation reports with quality scores, confidence reports, audit reports, remediation recommendations.

The Feedback Loops

FCC is not a linear pipeline. Critique phase personas feed findings back to both Find and Create phases:

graph TD
    RC[RC: Research Crafter] -->|research_inventory| BC[BC: Blueprint Crafter]
    BC -->|blueprints_specs| DE[DE: Documentation Evangelist]
    DE -->|publish_handoff| RB[RB: Runbook Crafter]
    DE -->|publish_handoff| UG[UG: User Guide Crafter]
    DE -.->|standards_edits| BC
    RB -.->|operational_feedback| BC
    RB -.->|operational_findings| RC
    UG -.->|usability_feedback| BC
    UG -.->|user_feedback| RC

Dashed lines represent feedback edges. When the Runbook Crafter discovers that the blueprint is missing error handling for API timeouts, that finding flows back to both the Research Crafter (to update the knowledge base) and the Blueprint Crafter (to update the specification). This is not optional polish -- it is how FCC iteratively improves quality.

The Persona Model

FCC uses 24 specialized AI agent personas organized into 5 categories:

Category	Count	Personas	Purpose
Core	5	RC, BC, DE, RB, UG	The essential FCC cycle
Integration	7	CIA, UMC, STE, TS, BV, RIC, GCA	Specialized capabilities
Governance	3	DGS, PTE, AMS	Data governance, privacy, fact-checking
Stakeholder	5	CO, SMC, EC, RS, SCP	Coordination, metrics, publishing
Champion	4	RCHM, BCHM, UGCH, RBCH	Elevated orchestrators

Each persona is defined by a 10-component R.I.S.C.E.A.R. specification that captures its Role, Inputs, Style, Constraints, Expected Output, Archetype, Responsibilities, Role Skills, Role Collaborators, and Role Adoption Checklist. This is covered in detail in the next tutorial.

What Makes FCC Different

Traditional Approach	FCC Approach
Ad hoc research	Structured Find phase with specialized research personas
Generalist writers	Specialized Create-phase personas for each artifact type
Manual review	Automated quality gates with configurable thresholds
No governance	Built-in compliance auditing, privacy classification, fact-checking
No feedback loops	Structured critique-to-find and critique-to-create feedback edges
Manual coordination	Champion personas orchestrating teams of specialists
One-size-fits-all	Three workflow graphs scaling from 5 to 24 personas

Running FCC

FCC operates through a simulation engine that propagates messages through a workflow graph. You can run simulations in two modes:

• Deterministic (mock) mode: Predictable message passing for testing and validation
• AI-powered mode: Each persona uses an LLM (Anthropic Claude or OpenAI) to generate contextual responses

Both modes produce trace output -- a JSON record of every message, every persona action, and every handoff in the workflow.

Next Steps

Now that you understand what FCC is and why it exists, proceed to:

• Understanding Personas -- Deep-dive into the persona specification model
• The FCC Cycle -- Detailed walkthrough of Find, Create, and Critique phases