Why flat multi-agent design failed: a 3-level hierarchy from a hospital ops system
Repulsive_Sugar_5252 · reddit · 2026-09-23
The author building Hospilot, a multi-agent orchestration system for hospital operations, explains why their early flat design (goal → agents) collapsed on a real domain: "bed management" bundles genuinely distinct concerns — availability checks, ventilation tracking, isolation rules, cascading bed effects — that don't belong in one blob of logic.
The redesign uses a 3-level hierarchy:
- agent = namespace/domain (bed management, ICU ops, staffing)
- sub-agent = a coherent slice of that domain's responsibility
- task = atomic unit of work with declared outputs
The planner routes goals at the agent level, while the real decision of which sub-agents fire and in what order happens one level down. Adding capability means adding a sub-agent or task without touching upstream dependents. Granularity matters: task-level was too fine for the planner to reason efficiently over hundreds of tasks, agent-level alone too coarse.
The post closes by asking how others choose hierarchy depth in multi-agent systems.
More from coding & agent
- Dev uses Opus 5.5 with Lean to formally verify Claude Agent SDK, shipping 16 bug-fix PRs — jimmykoppel · 2026-09-23
- Rogo CEO names memory compaction as the key unsolved problem for enterprise agents — rohanpaul_ai · 2026-09-23
- Graph Engineering: Building Reliable AI Agent Systems as Explicit Task Graphs — Pavan_Belagatti · 2026-09-23
- Have Your Coding Agent Attach Flame Graphs to Every PR It Opens — DanielLockyer · 2026-09-23
- Early hands-on: Sol 6 shows strength at goal-driven tasks, dev lets it run overnight — gregmushen · 2026-09-23
- Theorem says Lean-verified AI sandboxes are months away, at 1-30KB of proofs verified per hour — ctjlewis · 2026-09-23