Two different things
Automation
Software that follows instructions. A script that logs into a portal. A connection between two systems. A job that runs at 6am. A web app with the right screens. It has been around for decades, it is reliable, and it is what most office tasks actually need.
- Follows rules you can write down
- Same input, same result, every time
- Runs on a schedule or at the press of a button
- Easy to check: it either did the steps or it didn't
- Cheap to run for years
AI
Software that handles what rules can't describe. It reads a letter that arrives in any format. It understands a caller asking a question in their own words. It sorts a document when the naming rules aren't clean. It is useful for a narrow set of tasks, and it needs more care than automation does.
- Handles input that has no fixed format
- Makes a judgment call where rules run out
- Reads, listens, summarizes, drafts
- Needs a person checking the edges at first
- Costs a little every time it runs
Plain automation handles most repetitive work. That is scheduled scripts, system-to-system connections, and web apps with the right screens. We use newer tools only where they clearly earn their place, and we tell you when we do.
Where AI earns its place is narrow and specific. Voice agents for routine calls. Extraction from unstructured emails and documents. Classification where the rules are not clean. That is most of the list.
A test you can run on your own task
Write the steps on an index card and imagine handing it to a new hire on their first morning. If they could do the task from the card alone, without asking anyone, it is automation. Log in here, search for this, copy that field into this column. A rule is a rule, and software follows rules better than people do.
If the new hire would have to read something and decide what it means, listen to someone and work out what they want, or look at a document and judge where it belongs, the card runs out. That gap is where AI might belong, and only that gap. The rest of the task is still automation.
Most tasks fail the test in one small place and pass everywhere else. That is why most of what we build is ordinary software with, at most, one narrow piece of AI inside it.
Where common tasks land
The two systems just need to be connected. Once they are, the second entry happens on its own.
The steps never change. Log in, search, read a field, write it down.
The sources and the layout are known. The report can assemble itself before anyone is in.
If the sender or subject tells you where it goes, that is a rule. If a person has to open it and read it, that is where AI earns its place.
A form or a template is a rule. Free text written differently by every sender needs extraction.
Callers do not follow a script. Understanding what someone is asking, in their own words, is judgment.
How we decide, and how we tell you
We start with the work, not the technology. We sit with the task as it is actually done and write down the steps. Most of the time the steps are all there, and plain automation is the whole answer. It is cheaper to build, cheaper to run, and it does the same thing every day without anyone watching it.
When the steps run out, because a person has to read something or make a call, we use AI for exactly that part and nothing else. We tell you where the line is. Every task page on this site says whether it is plain automation, AI, or a mix, and the quote says the same. We build in a place for a person to check the results until you are comfortable with them.
What we won't do is put AI into a job that a rule could do. It costs more, it is harder to trust, and it gives you a harder problem than the one you started with.
Two examples where it does belong: routine inbound calls and filing documents when the naming rules aren't clean. One where it does not: re-keying data between two systems.