AGT-03 · Planner patterns: choose, build, tune and set expectations¶
Teams build bespoke planners for problems a simple loop or workflow already solves, and clients expect more autonomy than the design can deliver. When plans go wrong there is no shared way to see why or what it costs.
- Who reads it: GenAI engineers and solution architects designing agent reasoning, and engineers supporting it in production
- Outcome: The cheapest planner that meets the quality bar, with known cost per task
- Pack owner: James (catalog owner role: Agent Lead)
Master catalog entry¶
- Catalog name: Planner Pattern Catalog (P2)
- Why it matters: Sets realistic client expectations on what autonomy actually delivers
- What it removes: Stops teams building a bespoke planner for a solved problem
- Stack: LangGraph, ReAct, plan-and-execute, reflexion patterns
- Maturity target: Level 2 — five planners with cost and reliability trade-offs
Playbooks¶
| ID | Story | Title | Kind | Complexity | Urgency | Status | Owner | Planned publish |
|---|---|---|---|---|---|---|---|---|
| PB-AGT-03-01 | AGT-03-S1 | Choose a planner pattern: workflow, ReAct, plan-and-execute, reflexion or orchestrator-worker | Decision guide | Medium | 2 | draft | James | 2026-12-11 |
| PB-AGT-03-02 | AGT-03-S2 | Implement plan-and-execute with bounded replanning in LangGraph | How-to | Medium | 2 | draft | Sanjay Sarvesh | 2026-12-29 |
| PB-AGT-03-03 | AGT-03-S3 | Investigate bad plans, replan loops and planner cost spikes | Runbook | Medium | 4 | draft | Sanjay Sarvesh | 2026-12-11 |
| PB-AGT-03-04 | AGT-03-S4 | Planner and prompt change regression checklist | Checklist | Low | 3 | draft | Sanjay Sarvesh | 2026-12-21 |