Planning Infrastructure
How does work get scheduled, tracked, and measured across agents?
Three subsystems compose into one scheduling loop. The planning DB tracks what to do. The agent system tracks when to do it. The comms system tracks who said what.
System Overview
Entity Relationships
DREAM vs REALITY
Entity Relationships
| Entity | Table | Owns | Purpose |
|---|---|---|---|
| Project | planning_projects | Plans | Groups related plans — one per initiative |
| Plan | planning_plans | Phases, Tasks | One unit of tracked work, created from template |
| Phase | planning_phases | Tasks | Groups tasks into sequential stages |
| Task | planning_tasks | Steps, Acceptance Criteria | One unit of work with pass/fail gate |
| Session | planning_sessions | — | Records agent work windows with duration |
| Progress Event | planning_progress_events | — | Append-only log for crash recovery |
| Agent | agent_profiles | TimeMindPlan | Identity — AI or Human, with capabilities |
| TimeMindPlan | agent_time_mind_plans | Timeblocks | Weekly schedule with energy allocation |
| Timeblock | agent_timeblocks | — | Calendar slot: when, what, which archetype |
DREAM vs REALITY
Every timeblock has a blockType. DREAM blocks are the plan. REALITY blocks record what happened.
| DREAM | REALITY | Signal |
|---|---|---|
| Mon 08:00 SCAN | Mon 08:00 SCAN | On-intent |
| Tue 09:00 CREATE | Tue 11:00 CREATE | Partial — started late |
| Thu 08:00 DISTRIBUTE | (missing) | Off-intent — skipped |
Comparing DREAM and REALITY at the end of each cycle is the MEASURE phase. The gap between setpoint and gauge reading is the feedback signal. This is the same pattern that runs through the feedback loop architecture at every scale.
Template Lifecycle
Templates compound. Each completed plan can trigger the legacy rule — one concrete improvement to the template based on real experience.
Agents in the System
Each workflow domain has an agent that owns its timeblocks:
| Agent | Domain | Strategic Bet | Archetype |
|---|---|---|---|
| content-marketing | Content calendar | distribution | Varies by phase |
| orchestrator | Cross-team coordination | — | — |
| sales-dev | Business development | — | — |
| test-engineer | Verification | — | — |
Agents are universal — AI and Human share the same profile schema. The agentType field (AI/Human) is the only distinction. Both can own timeblocks, both can post to comms channels.
Connection to A&ID
This planning infrastructure maps to Agent and Instrument Diagrams. In A&ID notation:
The planning infrastructure IS an A&ID implemented in Drizzle tables and CLI commands.
Context
- Software Architecture — The broader technical stack
- Hexagonal Architecture — Where planning repositories sit in the hex layers
- Tight Five Loops — DREAM vs REALITY is the universal feedback pattern
- A&ID Templates — P&ID translated to agent systems
- Work Charts — The business context for AI-augmented workflows
- Content Calendar — The process that runs on this infrastructure
Questions
How do you know when a planning template has matured enough that it no longer needs the legacy rule?
- What signals distinguish a template that's stabilized from one that's stagnated?
- When should a REALITY timeblock that consistently deviates from DREAM trigger a template change rather than a discipline change?
- At what point does the overhead of tracking (plans, timeblocks, comms) exceed the value of the feedback it provides?