← Advanced learning mapbridge lesson · act

Workflow, copilot, or agent?

Distinguish a fixed process, a person-facing assistant, and a bounded system that chooses actions.

  • 10 minutes
  • foundation
  • Reviewed 2026-07-16
01

What is it?

A workflow executes designed steps. A copilot proposes or drafts while a person leads. An agent observes context and selects among permitted actions. Orchestration coordinates these parts across time.

02

Why does it matter?

Clear labels prevent needless autonomy. The simplest useful pattern is usually easier to test, secure, and operate.

03

The mental model

A workflow follows a defined path. A copilot assists a person. An agent can decide which action to take within defined boundaries.

04

A simple example

A copilot drafts a reply, a person approves it, and a workflow sends it. An agent is justified only if it must choose which approved tools to use for varied requests.

05

What it is not

A chatbot is not automatically an agent, and adding a model call does not make an entire workflow agentic.

06

Learn this first

These ideas make the lesson easier to place.

  • Deterministic versus AI decisions
07

Your first 60 minutes

Use one focused hour to make the idea concrete.

  • Write one task goal and its possible paths.
  • Decide who leads: workflow, person, or agent.
  • List the actions and boundaries before choosing technology.
08

Build this first

Create three versions of the same support scenario: fixed routing, copilot-assisted response, and bounded tool-using agent.

09

When not to use it

Avoid an agent when the path is known, the action is irreversible, the evidence is weak, or operators cannot observe and stop it.

10

What to learn next

Learn structured outputs, tool calling, and the agent loop.