Stop Building Dev Tools. The Real Money Is in the Boring Industries.
Go to any hackathon, demo day or launch feed right now and count how many of the projects are an AI tool for developers. A better code review bot. A smarter terminal. An agent that writes your tests. A framework for orchestrating other frameworks.
I am not throwing stones from a distance here. I have built these. It is the most natural thing in the world to build, because you are a developer, you understand the pain intimately, and you can validate the idea by asking the person sitting next to you.
That last part is exactly the problem.
You Are Building for the Worst Customer on Earth
Think about who you are actually selling to when you build a developer tool.
They are the most technically literate buyer alive, which means they can build a worse version of your product in a weekend and will genuinely consider it. They are the most price sensitive buyer alive, because they have been trained for twenty years to expect their tooling to be free and open source. They have close to zero switching costs, so the loyalty you earn on Monday evaporates when something shinier lands on Tuesday.
And above all of that, you are building in the one category that the model labs themselves are actively shipping into. Your entire product can turn into a bullet point in somebody's release notes, and there is nothing you can do about it because you were always standing on their infrastructure.1
Meanwhile the actual economy is sitting right there, running on spreadsheets and PDFs and phone calls, largely untouched.
What Is Actually Happening in the Boring Industries
I now work on a vertical AI platform in commercial real estate lending, and the thing that struck me in the first month was not how hard the problems were. It was how unglamorous and how enormous they were at the same time.
In real estate, a deal arrives as a PDF offering memorandum and a rent roll in whatever format the broker felt like exporting. Somebody, usually a very expensive and very tired analyst, retypes it into a model. In private equity, diligence means a data room of several thousand documents that a team reads through under a deadline, hunting for the three clauses that matter. In manufacturing, quality inspection is a person looking at parts, and the maintenance schedule lives in an ERP system that was configured in 2004 by a contractor who has since retired.
None of this is intellectually thrilling. All of it is a direct, expensive, load-bearing use of human hours. When you automate two analysts out of a forty hour week in private equity, you are not saving somebody a mildly annoying afternoon. You are removing a real line item that already exists in a real budget, from a firm whose margins are large enough that they will happily pay you a serious fraction of what they save.2
That is the difference. In a dev tool, you are trying to create a budget line that did not previously exist. In a vertical, you are competing against a salary that is already being paid.
Painkillers and Vitamins
The old framing still holds up better than almost anything else in startup advice. A vitamin is nice to have and improves your life at the margin. A painkiller addresses something that hurts right now, and people pay for painkillers without a procurement cycle.3
Here is the test I use now, and it is uncomfortable to apply to your own idea. If your customer stopped using your product tomorrow, does their job get measurably worse, or do they just go back to how things were and barely notice?
Most of what gets built in our circles fails that test. We build things we personally find cool, automate the workflows that we personally find irritating, and then perform the motions of running a startup around it. Landing page, waitlist, launch post, a logo that looks like every other logo. It has the shape of a company without the substance of one, because at no point did anyone with money and a problem ask for it.
You cannot find a painkiller by introspection. You can only find one by going and looking at somebody else's job.
Why Nobody Wants To Do This
If the opportunity is this obvious, the fair question is why the smartest people you know are still building the seventeenth agent framework.
Because the vertical path is genuinely unpleasant at the start. You have to learn an industry you have no interest in. You have to understand what a cap rate is, or how a debt schedule works, or why a plant manager cares far more about unplanned downtime than about your model's benchmark scores. You cannot ship on Friday and get a thousand GitHub stars by Sunday. Your first customers are found through relationships and introductions, not through a launch post, and the sales cycle is measured in months.
But sit with that for a second, because it is the entire point. Every one of those obstacles is a wall that keeps other engineers out.4 Technical difficulty is not a moat anymore, since the model that powers your product also powers everyone else's. Domain knowledge, earned slowly by sitting next to people who do the work, is one of the few moats left standing.
The incumbents in these industries will not stop you either. They are not software companies, they will not out-build you, and the ones who could buy their way in are usually too busy defending margins to notice a small team automating one workflow at the bottom of the market.5
How To Actually Start
Pick an industry where a lot of money moves and very little software exists. Then go and find someone who works in it and ask them to walk you through their week, hour by hour, without pitching anything.
You are listening for one specific thing: the task that eats an absurd number of expensive human hours and produces an output that is fundamentally structured. Reading documents and pulling out fields. Reconciling two systems that disagree. Turning a messy input into a clean format so that the next person down the line can do the real work. That shape of problem is exactly what these models are good at, and it is sitting in every one of these industries in large quantities.
Then charge for the outcome rather than for seats. If you save a firm four hundred analyst hours a quarter, price against that number and not against how many people happen to log in.
Find a design partner who actually has the problem before you write meaningful code. If you cannot find one person willing to give you their time and their real documents, you have discovered something valuable early: the pain was not as sharp as you assumed.
The Uncomfortable Bit
There is a version of this decade where a generation of talented engineers spends it building slightly better tools for each other, in the single most competitive category in software, while the largest industries in the world continue running on manual data entry.
There is another version where some of us go and learn what a rent roll is.
The second one is less fun to post about and worth substantially more, both in what you can build and in what actually changes for the people doing the work. Go find something that hurts, and fix that.