Build a Team
A single agent hits its limits fast. Real systems need teams — agents with distinct roles that complement each other. This lesson teaches you how to compose agent teams, which roles are essential for common use cases, and what happens when you pick the wrong team.
Why Teams, Not Solo Agents?
A single agent with 30 tools, 10 goals, and 5 different responsibilities will underperform a team of 3 specialized agents every time. Why?
LLMs perform better with fewer tools and a clear role. A "content writer" agent with 5 writing tools will produce better output than a "do-everything" agent with 30 tools that happens to also write content.
The agent that writes content should not also be the one that publishes it. If the writer crashes, the publisher keeps working on queued content. Each agent can fail independently without bringing down the whole system.
An editor agent reviewing a writer agent's output catches errors the writer would never catch on its own. An agent checking its own work is like a student grading their own exam — a separate reviewer is always better.
The Eight Core Roles
Most agent teams draw from these eight archetypes. Not every team needs all eight — the art is choosing the right subset for your use case:
Creates content — blog posts, emails, reports, social copy. The producer.
Reviews, fact-checks, and improves. Quality gate that prevents errors from reaching production.
Deploys content to websites and platforms. Without it, content sits in drafts forever.
Watches systems for errors. The first to know when something breaks. Essential for reliability.
Enforces rules, checks compliance, validates actions. The safety net for the whole system.
Sends alerts via email, Slack, or SMS. Keeps humans in the loop when agents act.
Manages timing, cron jobs, and queues. Ensures tasks run at the right time in the right order.
Analyzes metrics and generates insights. Turns raw data into actionable intelligence.
Team Recipes
Three proven team compositions for common use cases:
Writer creates, Editor quality-gates, Publisher deploys, Notifier confirms. Pipeline orchestration (Lesson 5).
Monitor detects problems. Guardian validates fix actions. Scheduler handles timing. Notifier alerts humans. Supervisor orchestration.
Scheduler triggers on cron. Analyst processes data. Monitor watches health. Notifier reports results. Pipeline + supervisor hybrid.
Scaling Teams: When to Add Agents
Adding agents to a team is not always the answer. Here is when adding a new agent is the right call — and when it is not:
An existing agent has too many tools (15+). A new responsibility has different failure modes. You need independent scaling for a specific task. Two roles should never share state or permissions.
A simple function call would suffice. The "agent" would only have one tool. Adding it creates unnecessary coordination overhead. The existing agent handles the task fine — you are just over-engineering.
This lesson is for Pro members
Unlock all 520+ lessons across 52 courses with Academy Pro.
Already a member? Sign in to access your lessons.