The useful distinction is not the product label. It is how much of the work the system coordinates. A chat interface can connect to tools, and an automation can start with a manual click. What matters is what happens after the request.
What does an assistant do?
A basic assistant responds to a person’s request. Somebody describes what they need, reviews the answer and decides what to do next. Some assistants can also use tools or run scheduled tasks, so the interface alone does not tell you how the process works.
That can improve drafting, summarising and analysis substantially. But if the next step still depends on someone copying the answer into another system, that handover has not yet been automated.
What does an automation do?
An automation coordinates an agreed sequence of steps. A document arrives, a form is submitted, a date is reached or a person starts a run. The system applies the rules and produces a result, with approval or exception handling where needed.
Ask who collects the input, checks it, moves it and confirms the outcome. Those responsibilities describe the difference more reliably than asking whether the tool has a chat window.
Why does that difference matter so much?
Consider an illustrative task that takes ten minutes and arrives eight times a day: eighty minutes of direct work. If an assistant reduces each task to six minutes, that releases thirty-two minutes. Context switching may add further cost, but it should be measured rather than assumed.
A connected workflow could also remove copying and routing. It will still have review time, exceptions and maintenance. Compare the complete process, not just the speed of the generated answer.
Where do assistants genuinely help?
They are useful for preparing a meeting, drafting a difficult reply, exploring ideas or turning notes into a clear summary. For high-stakes material such as contracts, a draft or summary supports qualified review; it does not replace it.
For irregular, low-volume work, keeping a person in control may be the simplest and most effective design. Repetition, cost and risk determine whether building a wider workflow is worthwhile.
Where are assistants used as a substitute for a process?
In the places where volume is high and the input is regular. Somebody pasting invoice text into a chat window forty times a day is doing manual labour with a faster tool.
That pattern is a useful signal. The employee’s prompt and corrections provide examples of what works, but they are not yet a complete specification for unattended operation.
Can one become the other?
A well-tested prompt can become part of a workflow specification. Add the input sources, access rules, required output, validation checks and examples that should stop for human review.
The remaining work depends on the systems involved. Integration access, duplicate handling, testing and monitoring can take more effort than the prompt itself. Start with one bounded handover and compare the result with the existing process.
What is the question to ask?
Ask what people repeatedly do before and after using the assistant. Record frequency, time and the consequence of an error. That reveals whether better training, a simpler template or a connected workflow is the right next step.
