"Not every task needs a genius that runs wild. Often the best 'agent' is the simplest one that reliably gets the job done."
Learning Objectives
By the end of this module, you will be able to:
- Recognize the common types of agents
- Match each type to the right kind of task
- Understand why simpler is often better
1. A Spectrum of Agents
"Agent" isn't one thing — it's a spectrum, from simple and predictable to complex and autonomous. Here are the types you'll meet:
| Type | What it is | Good for |
|---|---|---|
| Simple tool-using | An LLM that can call one or two tools | "Look this up and summarize it" |
| RAG agent | Looks up your documents before answering | Answering from a company handbook |
| Workflow agent | Follows a fixed set of steps (mostly scripted) | Predictable, repeatable processes |
| Autonomous agent | Plans its own steps freely toward a goal | Open-ended, hard-to-script tasks |
| Multi-agent system | Several agents with roles, working together | Big tasks split across "specialists" |
| Human-in-the-loop | Pauses for a person to approve key steps | Anything risky or high-stakes |
Read the table top to bottom and you'll notice a trend: each row hands the AI a little more freedom and a little less predictability. The top rows are easy to test and trust; the bottom rows are more capable but need more care.
Key idea: More autonomy is not always better. Autonomy adds power and unpredictability. The skill is choosing the least complex option that reliably solves your problem.
2. Workflows vs. Autonomous Agents
This is the most useful distinction for beginners:
- A workflow follows steps you defined ("do A, then B, then C"). It's predictable and easy to debug.
- An autonomous agent decides the steps itself. It's flexible but harder to predict and test.
Reputable engineering guidance is consistent: start with a workflow, and only reach for full autonomy when the task genuinely can't be scripted.
Explain like I'm new: A workflow is like a train on tracks — it can only go where the rails lead, which is exactly why you always know where it'll end up. An autonomous agent is like a car with a driver: it can reach places a train never could, but you're trusting the driver's judgment at every turn. Use the train when the tracks already go where you need.
3. Multi-Agent Systems
Sometimes one agent isn't enough. In a multi-agent system, you give each agent a role — like a "researcher," a "writer," and an "editor" — and they collaborate, passing work between them. It's powerful for big tasks, but every extra agent adds cost and complexity.
Real-world use case: Imagine producing a market report. A researcher agent gathers data, hands it to a writer agent that drafts the report, which passes it to an editor agent that checks facts and polishes tone. Splitting the work like a small team can raise quality — but it also means three sets of costs, three chances for miscommunication, and three things to debug when the output is wrong.
Common mistake: Jumping straight to a fleet of autonomous multi-agents for a task a simple workflow could handle. It's slower, pricier, and harder to trust. Build the simplest thing that works; add agents only when you feel a real limit.
4. How to Choose
When you're not sure which type fits, walk down this short list and stop at the first "yes":
- Does the answer live in your documents? Start with a RAG agent.
- Are the steps always the same? Use a workflow.
- Is the action risky (money, deletion, mass email)? Add human-in-the-loop on top of whatever you pick.
- Are the steps unpredictable and hard to script? Now consider an autonomous agent.
- Is the task genuinely too big for one agent? Only then reach for a multi-agent system.
The order is deliberate: it moves from simplest and safest toward most powerful and least predictable. Most beginner projects never make it past the first two bullets — and that's a sign you're doing it right, not a shortcoming. Complexity is a cost you pay only when a simpler design has visibly failed you.
Key Takeaway: Agents range from simple tool-users and RAG agents to workflows, autonomous agents, multi-agent teams, and human-in-the-loop setups. Workflows are predictable; autonomy is flexible but risky. Pick the simplest type that reliably solves the task — and add complexity only when you must.