Automation follows the path you drew. An agent decides which path to take. Knowing which one you actually need changes what you build.
Executive Summary
The words get used as if they mean the same thing. They do not. Automation and agents solve different problems. Treating them as interchangeable leads to tools that disappoint in both directions: automation asked to think, and agents handed work that never needed judgement.
The distinction is simple once you see it, and it changes how you decide where to spend.
What automation actually does
Automation runs a path you defined in advance. When this happens, do that. Move the file, send the email, update the record, flag the exception. It is fast, reliable, and predictable, because you told it exactly what to do.
Its strength is also its limit. Automation does not handle situations you did not foresee. Show it something off the script and it stops, or worse, it carries on regardless.
What an agent does differently
An agent is given a goal rather than a path. It works out the steps, uses the tools available to it, and adapts when the situation is not what it expected.
Ask automation to "process this invoice" and it follows the invoice rules you wrote. Ask an agent and it can read a document it has not seen before, decide what kind of thing it is, and route it accordingly. The difference is judgement, applied within limits you set.
Why the confusion costs money
Pick the wrong one and you pay for it twice.
Use an agent where simple automation would do, and you have added cost, unpredictability, and a thing that needs watching. All to do a job a rule could have handled. Use automation where the work genuinely varies, and you get a brittle process that breaks the moment reality stops matching your assumptions.
Most disappointment with AI in operations traces back to this mismatch, not to the technology itself.
A simple way to decide
Ask one question about the task. Does it vary in ways you cannot fully predict.
If the answer is no, the steps are knowable and stable, automation is the right tool. It will be cheaper, faster, and easier to trust. If the answer is yes, the task needs judgement on inputs you cannot script in advance, that is where an agent earns its place.
Most businesses need far more of the first than the second. That is not a limitation. It is good design.
Where they work together
The strongest setups use both. An agent handles the part that needs judgement, then hands off to automation for the steps that are set.
The agent decides what kind of request this is. Automation then runs the known process for that kind. You get judgement where you need it and reliability everywhere else, which is usually exactly what the work calls for.
The goal was never to use the most advanced tool. It was to match the tool to the shape of the work.