AI Agents Engineering · What an Agent Actually Is · Lesson 1 of 3
Agent vs. single call vs. workflow
An agent is an LLM running in a loop, using tools to pursue a goal over multiple steps, deciding its own next action. A single LLM call takes input, returns output, once. An agent observes, decides, acts, and repeats until the goal is met or a limit is hit.
- Single call: deterministic path, one shot. Cheap, fast, predictable.
- Workflow: a fixed chain of LLM calls / tools (known steps, coded by you). Still predictable and testable.
- Agent: dynamic path — the model chooses which tools to call and when, and decides when it is done. Powerful but harder to control.
The senior instinctPrefer the simplest thing that works. Reach for an agent only when the path is genuinely dynamic and cannot be pre-scripted. Every autonomous step adds latency, cost, and a new way to fail.
A useful test: if you can draw the flowchart in advance, build a workflow. If the branching depends on what the model discovers at runtime, you may need an agent.
◆ Lock it in
- Agent = LLM in a loop with tools, choosing its own next action and stop point.
- Ladder of control: single call → workflow → agent; climb only as needed.
- If you can draw the flowchart up front, it is a workflow — agency is a cost, not a default.
Feynman drill — say it out loudExplain the difference between a fixed workflow and an agent, with one concrete example of each.
Step 1 rate your confidence · Step 2 pick your answer
When is a fully autonomous agent the RIGHT choice over a fixed workflow?
Step 1 — how sure are you?