AI Development: Architecture

Category: AI Development Last Updated: 2026-01-26

Overview

AI-assisted development system including Agent Farm orchestration, multi-LLM consultation, and development protocols.

System Architecture

Architect (Claude)
    ↓
Agent Farm Dashboard
    ↓
Builder Agents (Claude)
    ↓
Code Repository

Key Components

Agent Farm (af)

  • Purpose: Orchestrate multiple agents, manage worktrees
  • Commands: start, spawn, status, send, cleanup
  • Implementation:

Consult Tool

  • Purpose: Multi-LLM review and validation
  • Reviewers: Gemini, Codex, Claude
  • Review Types: spec-review, plan-review, impl-review, pr-ready, integration-review

Codev CLI

  • Purpose: Project management and protocol execution
  • Commands: init, adopt, update, doctor

Protocols

SPIDER

  • Phases: Specify, Plan, Implement, Defend, Evaluate, Review
  • Use Case: Multi-phase features with formal review

TICK

  • Phases: Task, Implement, Check, Kontinue
  • Use Case: Small, well-defined tasks (<300 lines)

EXPERIMENT

  • Use Case: Research and exploration

MAINTAIN

  • Use Case: Codebase maintenance and updates

Agent Roles

Architect

  • Specifications, planning, builder coordination, integration review

Builder

  • Implementation, testing, PR creation

Consultant

  • External review and validation

Data Flow

User Request
    ↓
Architect (Spec + Plan)
    ↓
Consult (3-way review)
    ↓
Builder (Implementation)
    ↓
Architect (Integration review)
    ↓
Merge & Deploy

See also: Workspace Architecture