Agent Architecture/Subagents & Agent Teams
Advanced10 min

AutoDream — Structured Planning Artifacts

AutoDream is a discovery tool that generates structured planning artifacts — architecture docs, task breakdowns, risk assessments — from a natural language description of what you want to build. It is the pre-session stage before you open a Claude Code session or spawn an agent team.

Quick Reference

  • AutoDream is a discovery tool, not an execution tool — it generates plans, not code
  • Outputs: component diagrams, dependency graphs, task lists with effort estimates, risk areas, suggested CLAUDE.md rules
  • Use AutoDream before greenfield projects or large agent team runs
  • AutoDream artifacts become the context: blocks in AGENT.md definitions
  • Shipped in Q1 2026 alongside Remote Control, Dispatch, and Channels
  • AutoDream also refers to the background memory consolidation system (distinct from planning)
  • Differs from Plan Mode: Plan Mode scopes a known task; AutoDream discovers what to build
  • Differs from /ultraplan: /ultraplan refines a plan collaboratively; AutoDream generates it from scratch

What AutoDream Does

Two Distinct Features Under One Name

AutoDream refers to two related but distinct features that shipped in Q1 2026: (1) the Structured Planning Artifact generator, which produces architecture docs and task breakdowns from natural language; and (2) the background Memory Consolidation system, which runs like REM sleep to prune and refresh Claude's memory files. This article covers both.

The planning artifact generator answers the question: 'I know what I want to build — how should I decompose it and who should build what?' You give AutoDream a natural language description of intent, and it produces the documents you would normally write by hand before opening a development session.

ToolStageInputOutput
AutoDreamPre-session discoveryNatural language intent descriptionArchitecture docs, task breakdown, risk assessment, CLAUDE.md suggestions
Plan ModeIn-session scopingTask with existing codebase contextImplementation plan for a known task
/ultraplanPre-execution refinementDrafted plan needing human reviewAnnotated, approved plan ready for execution