agentic-ai-engineering/site/blog/posts/why-multi-agent.md

2.1 KiB

Why One Agent Is Not Enough

June 3, 2026

A single agent, no matter how smart, hits three ceilings.

The Context Ceiling

One agent doing everything means one context window holding everything. File contents, database schemas, business logic, deployment configs — they all compete for the same limited space. A 100K token codebase leaves zero room for the actual task.

The Capability Ceiling

A generalist agent is mediocre at everything. A specialized agent — backend dev, security reviewer, UI designer — outperforms the generalist in its domain by a wide margin. You wouldn't ask your frontend dev to write SQL migrations. Why ask your agent to?

The Reliability Ceiling

One agent failing = everything fails. Multi-agent systems degrade gracefully. If the reviewer agent crashes, the builder keeps working. If the planner goes down, deployed agents continue running.

The Solution: Depth-2 Delegation

The production-proven pattern:

User
  |
  v
Orchestrator (Opus-level, thinking only)
  |
  +-- Planning Team Lead
  |     +-- Product Manager (worker)
  |     +-- UX Researcher (worker)
  |
  +-- Engineering Team Lead
  |     +-- Frontend Dev (worker)
  |     +-- Backend Dev (worker)
  |
  +-- Validation Team Lead
        +-- QA Engineer (worker)
        +-- Security Reviewer (worker)

Leads think, plan, and delegate. Workers execute. The orchestrator never touches code.

When to Go Multi-Agent

You need multiple agents when any of these are true:

  • Different expertise required — Your task needs a planner, a coder, a reviewer, and a security auditor
  • Parallel work possible — Multiple sub-tasks can run simultaneously
  • Failure cost is high — You want a second agent to catch mistakes before they ship
  • Context exceeds one window — Split across specialized agents

If none of these are true, a single well-configured agent is simpler and cheaper.


This is an excerpt from Module 4 of the Agentic Engineering Course. The full module includes runnable lab code for building multi-agent systems with domain locking, mental models, and P2P communication.