//pragmatic leaders

Multi-Agent Orchestration — fan-out, swarms, and the cost of doing both badlyfeatured

The instinct to "add another agent" almost always loses. Ten lessons on when fan-out is a multiplier, when it is a tax, and how to tell the difference before you sign the bill — built around the live Devin-vs-Anthropic debate and the lived PL parallel-triage incident.

Category

AI Native

Level

advanced

Length

9 lessons

For

PMs, founders, and engineers who are scoping or evaluating multi-agent systems and need to read the trade-offs before the swarm arrives

Start course →

Lessons

Framing

01The two-essay debate — Cognition vs. Anthropic on multi-agent systems6 min 03When fan-out is the right primitive — research, exploration, parallel scoring6 min

Patterns

04Swarm patterns — manager-worker, peer-to-peer, pipeline7 min 05Handoffs — the failure surface between agents6 min 06Shared state, locks, and the consistency problem at agent scale6 min 07Quota pooling and blast radius — the under-discussed cost of fan-out6 min

Lived lessons

08The PL parallel-triage 429 incident — anatomy of a multi-agent failure6 min

Operating multi-agent

09Debugging multi-agent systems — why it's hard and what compensates7 min

Capstone

10Capstone — write the design memo that argues for or against a second agent7 min