Capstone

Multi-Agent Systems · ~10 min

The Multi-Agent Decision Table

Eight lessons, one reflex: default to one agent, and add coordination only where the task structure forces it. Here's the whole course as a lookup table — and a mixed review to prove it stuck.

Why this, for you: multi-agent design is a sequence of small decisions, each with a default and a backfire. The table below maps the symptom you're staring at to the pattern that answers it — and, just as often, to "you don't need another agent." That second answer is the one this course exists to make automatic.

The through-line of the whole course: coordination is a cost you pay to buy independence or verification — never a free upgrade. Reach for the smallest structure the task actually demands.

1 Symptom → move

SymptomMoveFrom
A single agent struggles on a hard taskClimb the ladder one rung; reach multi-agent only when simpler levels stop being reliable — budget ~15× tokensL1
Work splits into independent directionsOrchestrator-worker — decompose, parallel workers with scoped tools, then synthesize (not concatenate)L2
High-stakes design needs the best of several attemptsFan-out N=3–5 with engineered diversity, then a synthesis agent + committee reviewL3
A child must disagree with the parentSpawn fresh — fork only to extend a load-bearing decision with siblings to amortizeL4
Information is lost between pipeline stagesTyped handoff — conclusions not transcripts; gate the first write with plan approvalL5
An agent declares "done" on broken workVerify-gated completion — independent read-only verifier on the critical path, fail closedL6
The system misbehaves and you want a bigger modelDiagnose the MAST bucket first — 41.8% of failures are design; rewrite prescriptive prompts as framework promptsL7
You need to ship a prompt/model change in productionRainbow deploy — N versions drain independently; roll back the whole tuple via a router changeL8

2 The one rule under all of it

Add an agent only to buy independence (parallel directions, an unbiased second opinion) or verification (a check the producer can't self-grade). Everything else is coordination cost: ~15× tokens, a new misalignment surface, and emergent sensitivity to every prompt word.

Part 1 was when and how: the complexity ladder gates the whole decision, and the two canonical shapes — orchestrator-worker (one task into subtasks) and fan-out synthesis (one task, N attempts) — both live or die on the synthesis step, never a concatenation. Part 2 was the contracts that hold a fleet together: fork-vs-fresh on each spawn, a typed handoff plus plan gate between stages, and admission control over every "done." Part 3 was operating the fleet: the failure taxonomy that tells you to suspect the design before the model, and rainbow deployment for shipping changes without breaking live sessions. One move, applied at each layer — pay for coordination only where the structure earns it.

Don't over-build the fleet

Every lesson had a backfire box, and they rhyme: most multi-agent configurations lose to a strong single agent. Fan-out on a routine task, a plan gate on a one-file edit, a verifier with no ground truth, a rainbow deploy for single-turn Q&A — each adds cost with no return. Reach for coordination when the task genuinely decomposes or a mistake is expensive, not on reflex.

↪ Your win: a multi-agent reflex

Mixed review — across all eight lessons

Question 1 · from L1The default reflex before reaching for multiple agents is to…

Question 2 · from L2An orchestrator that concatenates worker outputs adds…

Question 3 · from L4For a child that must reset bias and disagree, you spawn…

Question 4 · from L6In admission control, an ambiguous "done" claim resolves to…

Question 5 · from L8A deployable agent version should be rolled back as a tuple of…

You finished the course. Ask me to apply the decision table to a real multi-agent system of yours, draft an orchestrator prompt with effort budgets and a handoff schema, or design a verify-gated completion boundary with real ground truth. Or revisit any lesson — the patterns compound when you use them together.
✎ Feedback