Blog / Build with AI /
When automation is not worth it
I like automating things. It’s a small thrill to watch a script do in three seconds what used to take twenty minutes. That’s exactly why I don’t trust my own enthusiasm.
My argument is simple: automation is a cost you pay now to save time later, and plenty of tasks never repay it. With AI making scripts quicker to write, it’s tempting to automate everything. That’s a mistake worth catching early.
The sum nobody does
Before building anything, work out:
- How long does the task take each time?
- How often do I do it?
- How long will it take to build, test and document an automation?
- How long to maintain it?
If a task takes ten minutes, happens monthly, and the script takes an afternoon to get right, you’ll break even after about two years, and that’s if it never breaks. A well-known chart that circulates among programmers makes the same point: the time you can spend on automating something depends on how often you do the task and how much time you’d save.
AI changes the build time, not the whole sum. It’s quicker to get a first version, but you still pay for testing, fixing and keeping it alive.
Reason 1: tasks that change
If the process keeps changing, because the platform updates, the client alters the format or the rules shift, your script becomes a second job. A stable, repeated task is a good candidate. A shifting one is not.
Reason 2: tasks that need judgement
Choosing which pages to redirect, deciding if a comment is spam, writing to a client who’s upset: these look like routine work, but the value is in the thinking. Automating the surrounding mechanics can help. Automating the decision usually doesn’t.
Reason 3: the maintenance bill
Every automation is something else to monitor. It will fail one day, silently, because a login expired or a column was renamed. If nobody notices, you now have a process that appears to work and doesn’t. A manual task fails visibly because a human is doing it.
If you build it, add a way to know it ran: an email, a log entry, a count of items processed.
Reason 4: the risk is bigger than the saving
Scripts that edit live data, send customer emails or delete files can cost far more than the time they save when they go wrong. For those, I’d want a dry-run mode, a backup and a human check, and at that point the time saving shrinks.
Reason 5: you won’t remember how it works
Six months later, you open a script you didn’t fully understand even when it was new, written by a chatbot. It breaks. You can’t fix it. The ten minutes you saved each month cost you an evening of confusion. If you can’t explain it, document it, or ask someone to maintain it, it’s not finished.
The counter-argument
Fair objection: automation isn’t only about time. It’s also about consistency and removing boredom. A weekly checklist you always forget is worth automating even if the sum is close. Errors from tired humans count, and so does sanity.
I agree, and that changes the calculation. If a task is error-prone, regular and annoying, the saving is more than minutes. For the rest, I use a rule of thumb.
A rule of thumb that works for me
Do it by hand three times first. By the third time you know the steps and the awkward cases, and you can tell whether it’s really repeated. Then write the steps down. Often that checklist alone fixes most of the problem. Only then decide whether code is worth it.
This matches how I’d approach automating boring WordPress tasks with AI-written snippets: start with a task you understand fully, and not one you’re hoping the AI will figure out.
What to automate first
If you do decide to automate, pick the boring, stable, frequent ones. Examples a web admin might recognise: renaming a batch of image files to a consistent pattern, producing a weekly list of broken links to review, converting a client’s spreadsheet into the format a plugin needs, or creating the same set of pages for a new site. Each is clear, checkable and repeated.
Start with the smallest slice. Don’t automate the entire reporting process; automate the step where you copy figures between two sheets. If that works and saves time, extend it. If it doesn’t, you’ve lost an hour rather than a weekend.
Better than automating: eliminating
Sometimes the right move isn’t a script. It’s deleting the task. Does anyone read that weekly report? Does the spreadsheet need ten columns? Could a template remove the need to rewrite the same email? Removing work is cheaper than automating it.
Count your time honestly
When I do the sum, I try to include the unglamorous parts: reading error messages, re-running tests, writing the note that explains it, and the day it fails. Most people count only the writing. If your estimate for building is “an hour”, I’d double it and then add the maintenance. If it still looks worthwhile, you’ve got a good candidate.
A short checklist
- Does it happen often enough to matter?
- Is the process stable?
- Can I check it ran correctly?
- Is the downside of a mistake small, or protected by a backup?
- Could I explain it to someone else?
- Could I simply stop doing it?
If most answers are yes, go ahead. If not, a checklist is probably enough.
Related reading
- ChatGPT for builders: what it is best at
- Building an order export tool with AI
- Building a stock alert script with AI
Which task have you automated that, looking back, cost you more than it saved?