Automate the boring thing first
Everyone points AI at their most impressive problem. The right first project is the dull one — here is how to tell which is which.
When a company gets access to serious AI, the instinct is to point it at the most impressive problem. Don’t. Point it at the boring one first.
The boring one looks like this: a process that is already written down, follows rules, happens every week, and that nobody enjoys doing. Invoices re-keyed into a second system. A report assembled by copying numbers out of three places. Quotes checked against a price list by hand. Work that is tedious precisely because nothing about it is ever surprising.
That tedium is the qualification.
Why boring wins
Two reasons. First, you can verify it. You already know exactly what correct looks like, so when the software gets it wrong, you notice within a day — not three months in, when the error has compound interest. Second, its requirements are known. The impressive project — the assistant that answers customer questions, the dashboard that predicts something — has requirements nobody has written down yet. Automate it now and you are automating a guess. The boring process was specified years ago, by accident, by being done.
The page test
If a process can’t be described on one page, automating it isn’t the next step. Writing the page is.
The exercise is worth more than it sounds. Half the steps in an old process usually exist because of a limitation that no longer applies — a system that has since been replaced, a person who has since left, a rule that was actually a workaround. Describing the work forces those fossils to the surface, and the process gets simpler before any code touches it.
The description is the deliverable. The software just types it into a machine.
What the first project buys you
A boring first project is cheap tuition. It teaches you whether the tools fit your shop, where your data is actually clean and where it quietly isn’t, and how your team behaves when software does a job they used to own. None of those lessons are learnable in planning. All of them get more expensive to learn later.
And it banks a small, undeniable win. When the interesting project starts — and it should — it starts on rails that already work: data that has been through a machine once, a process somebody trusts, a team that has seen software succeed here before.
AI removed the cost of building things. It did nothing to remove the cost of being wrong about what to build. Boring work is where being wrong is cheap and the learning is fast. Start there, and the impressive project stops being a gamble.
Working on something this applies to? Tell us about it — thirty minutes, no pitch.