Adr 020 Agent Parallelization Strategy

EPGOAT Documentation - Architectural Decisions

ADR-020: Agent Parallelization for Content Work

Date: 2025-11-10 Status: ✅ Accepted Deciders: CTO (Claude Code AI), CEO Tags: #automation #efficiency #architecture


Context

CEO granted 25M token budget for comprehensive documentation work. Two approaches: 1. Sequential execution: Main conversation does all work 2. Parallel execution: Launch agents for independent work

Identified two parallelizable tasks: - Content Expansion (150+ expansion entries) - Legacy Migration (151 files to audit/migrate)


Decision

Launch 2 agents in parallel for independent content work while main conversation builds infrastructure:

Agent 1: Content Expansion - Create 150+ expansion entries for _STANDARDS, _CODEBASE, _WORKFLOWS - Autonomous, returns YAML files when complete

Agent 2: Legacy Migration - Audit all 151 legacy files - Migrate top 60, rewrite 40, archive 51 - Returns migration plan + migrated files

Main Conversation: Build infrastructure (hooks, ADRs, Cloudflare deployment)


Rationale

Why Parallel Agents?

  • Content expansion is independent of infrastructure work
  • Legacy migration doesn't depend on new system features
  • Saves 8-12 hours of sequential work
  • Maximizes 25M token budget utilization
  • All work happens simultaneously

Why NOT Sequential?

  • Would take 39-51 hours total (sequential)
  • Inefficient use of available resources
  • Agents idle while main does infrastructure work

Why 2 Agents (Not More)?

  • Only 2 truly independent large tasks identified
  • More agents = coordination overhead
  • Diminishing returns beyond 2

Consequences

Positive

  • ✅ Saved 8-12 hours of sequential work
  • ✅ Maximized resource utilization
  • ✅ Both agents completed successfully
  • ✅ High-quality outputs (150+ expansions, 42 files migrated)

Negative

  • ❌ Requires integration work after agents complete
  • ❌ Cannot monitor agent progress in real-time

Implementation

Execution: 1. Launched both agents simultaneously via parallel Task tool calls 2. Main conversation built hooks, living documents, ADRs 3. Agents returned complete outputs 4. Integration phase combines all work

Results: - Content Expansion Agent: ✅ 150+ expansions created - Legacy Migration Agent: ✅ 42 files migrated, plan created - Main: ✅ Hooks, living docs, ADRs created - Total time saved: ~10 hours


References