This is a question we get asked more than almost any other, usually by a slightly technical owner or someone with a developer friend, and it goes something like, why would I pay for an AI tool when I could just build my own with all these models lying around now. And we want to answer without the awkwardness because we're obviously on one side of it, we build and sell AI tools, so the cynical read is that of course we'll tell you to buy. So let us earn your trust by actually steelmanning the build case first, because it's more reasonable than the buy-side usually admits.
The build case is genuinely tempting right now, and not crazy. The models are cheap to access, the tutorials are everywhere, the barrier to wiring up a basic AI that does something has never been lower, and there's a real satisfaction and a real cost saving in owning the thing yourself rather than renting it monthly forever. For a business with some technical capability, the prospect of building exactly what you want, tuned precisely to your situation, with no subscription and no vendor to depend on, is legitimately appealing, and we'd be lying if we said the impulse was foolish. It isn't foolish. It's just usually a trap, and the trap is well disguised.
The reason it's usually a trap is hiding in the MIT research that's haunted this whole series, which found that companies building their own AI solutions tended to do worse than companies buying focused ones. That's a striking finding precisely because it runs against the build instinct, and the explanation is that the model was never the hard part. Standing up something that talks is easy now. The hard part is everything around it, the connection to your real data, keeping that current, the accuracy, the reliability, the handling of the weird edge cases, the not-falling-over when something changes, and that hard part is enormous and unglamorous and never finished, and it's exactly what the build fantasy underestimates.
We see this play out in a predictable arc that's almost painful to watch. The build starts strong, the demo works, the owner is thrilled, look what I made in a weekend. Then the real customers arrive with their real messy questions, and the edge cases pile up, and the thing that worked in the demo keeps stumbling in production. Then the underlying model gets updated and something breaks. Then the business changes and the data needs reconnecting. Then the developer friend who built it gets busy or leaves, and now there's a homegrown system nobody fully understands, gradually rotting, that the business has become quietly dependent on. The weekend project became an unpaid permanent job.
That last part is the cost the build case almost always omits, the maintenance, which isn't a one-time expense but a forever one. Buying a tool means someone else carries the burden of keeping it working as models change and the world shifts, which in a field moving this fast is a genuinely heavy and continuous burden. Building it yourself means you carry that burden, permanently, and it's the kind of burden that's invisible at the start and crushing over time, the slow accumulation of fixes and updates and edge cases that turns a clever project into an anchor. The subscription you were trying to avoid was, in part, paying someone to carry exactly this weight so you didn't have to.
There's also an opportunity cost that we think gets badly underweighted, which is what the building time could have been spent on instead. A small business owner's time and attention are the scarcest resources they have, and pouring them into building and maintaining an AI system is pouring them into something that is almost certainly not the actual business. The hours spent debugging your homegrown bot are hours not spent serving customers, growing the business, doing the thing you're actually good at and that actually makes money. Even if the build technically works, it may be the most expensive thing you ever made, just measured in the right currency, which is your attention.
We'd be fair and say there are real cases where building makes sense, because there are. If AI is genuinely core to your product, if you have real ongoing engineering capacity, if your needs are so unusual that nothing bought could fit, then building can be the right call, and we'd never tell such a business to buy out of dogma. The build-versus-buy answer isn't universal, it's situational, and the situation that favors building is real but much rarer than the build enthusiasts assume. For most small businesses, where AI customer service is a means to an end rather than the end itself, the situation points fairly clearly toward buy, and the enterprise data points the same way.
The deeper principle we keep landing on is that you should build the things that are your actual differentiation and buy the things that aren't. Your edge as a business is your work, your relationships, your judgment, the thing customers actually come to you for, and that's worth building and owning and obsessing over. The AI plumbing that answers routine questions is, for almost every business, not the differentiation, it's infrastructure, and you don't build your own infrastructure when good infrastructure can be bought, the same way you don't build your own payment processing or your own email server unless you have a very specific reason to. That isn't the same as saying every bought tool is equal, because most of them are all mouth and no memory, and the difference is whether the thing actually knows your business. But that is a question about which to buy, not about whether to build.
Where we think this shakes out, as the build tools get even easier and the temptation grows even stronger, is that more businesses are going to try building and more are going to discover the maintenance trap the hard way, the same way the enterprises did. The ease of starting will keep luring people in, and the difficulty of finishing and maintaining will keep them stuck, and the pattern the MIT research already found will probably just intensify, with the buyers quietly outperforming the builders not because buying is virtuous but because it lets the business stay focused on the business. The build fantasy is seductive precisely because the start is so easy and the cost is so deferred.
So our answer is that for most small businesses the build instinct is understandable, the start is genuinely easy, and the whole thing is usually a trap that costs far more than the subscription it was meant to avoid, mostly in time and maintenance and distraction. The question to ask isn't can I build this, because increasingly you can, it's whether building it is the best possible use of your scarcest resource, your own attention, and whether the thing you'd be building is actually your business or just plumbing for it. For most, the answer points toward buying the plumbing and building the business, which happens to be the side we're on, but which we'd believe even if we weren't.
