Build Boring

The instinct, when a small business decides to adopt AI, is to build the impressive thing. The customer-facing chatbot. The autonomous agent. The thing that looks good on a board slide and demos well to investors.

I want to argue, as gently as I can, that this instinct is wrong almost every time, and that the work that actually moves an SMB forward is quieter, less photogenic, and considerably more boring than the work that gets built.

Build boring. Then build boring again. Then, maybe, build something impressive — but only after you have learned what impressive costs to actually run.

A boring AI win, in my taxonomy, has three properties. First, it shaves time off a task someone on the team does every day. Second, that task is so mundane that nobody currently writes about it in their job description. Third, the people doing it would not have asked for the change because they had stopped noticing the friction.

Examples: turning a forty-minute weekly report into a four-minute one. Drafting the routine email that everyone sends fifteen times a day. Categorising the inbox before anyone reads it. Pulling the same five fields out of every PDF the business receives. Writing the first draft of the meeting summary the moment the call ends.

None of these are interesting. All of them are leverage.

Boring wins are passed over for three reasons, and they're all psychological.

They don't impress anyone. You cannot put "we reduced report-drafting time by thirty-five minutes a week" on the homepage. You can put "AI agent handles customer enquiries autonomously" on the homepage. Founders default to the homepage version because the homepage is what they think AI is for.

Nobody is asking for them. The people doing the boring tasks have been doing them for so long that the friction has gone invisible. They will not raise it at the all-hands. They will not put it on a wish list. The work has become just how things are done here, which is the most expensive sentence in business.

They feel beneath the technology. When you have a model that can write code and reason through paragraphs and pass professional exams, using it to format a spreadsheet feels like buying a Ferrari to drop the kids at school. The feeling is real. It is also wrong, because the Ferrari does the school run very well and you do the school run every day.

Sit with each of your highest-paid people for one hour. Not to ask them what AI should do. To watch them work. Specifically: watch them do the things they do on Monday morning, the things they do at five p.m. on Friday, and the things they do when they're tired.

In each session, you are looking for one thing: the task they perform that they have stopped narrating. The task that has dropped below their conscious attention and is now just muscle memory. That task is your boring win. It is also, almost certainly, costing the business an order of magnitude more than anyone realises, because it is being done by your most expensive people at the times they are least sharp.

If you find three of these per person — and you usually find more — you have your first hundred hours of AI work mapped out, and none of it involves an autonomous agent or a customer-facing anything.

Here's the part most businesses miss when they default to the impressive build.

Boring wins compound. Each one teaches the team how the model behaves, what it can be trusted with, where it breaks, and how to course-correct. Each one builds the internal muscle of deploying AI — which is the actual durable capability you're trying to develop, not any specific feature. By the time you've shipped your fifth boring win, your team is competent in a way that no demo project would have made them.

Impressive builds, by contrast, are usually one-shots. They consume a quarter of effort, ship in a blaze of internal attention, and then sit there. The team does not learn how to deploy because they only deployed one thing. The maintenance burden is large. The next AI project starts again from zero.

Boring is a strategy, not a fallback. It is the strategy that produces compounding capability, instead of one-off demos.

It takes a particular kind of courage for a founder to stand in front of an investor or a board and say we have shipped seven boring AI wins this quarter and our team is materially more capable. The narrative is unsatisfying. The press release does not write itself.

Do it anyway. The market punishes theatre as soon as it stops working, which it always does. The market rewards businesses that have actually become more capable, which is a quieter and longer-lasting story. Build boring, build relentlessly, and trust that the compounding will be visible to anyone who knows what they're looking at.