The AI Boom Only Counts If Small Businesses Get It

I think about AI as two different futures that happen to share a headline. In one, it becomes the most broadly enabling technology anyone has ever built: cheap enough, simple enough, and honest enough that a business of any size can use it to get meaningfully better at what it already does. In the other, it becomes the most entrenching technology yet, a force multiplier for whoever already has the money, the engineers, and the time to bend it into an advantage, while everyone else quietly falls behind.

Both futures are live right now. You can already see the entrenching one playing out in AI visibility, which is the part of the internet I spend most of my working life inside. Ask ChatGPT or Perplexity or Gemini which plumber to call, which bakery to trust, which accountant actually understands small business taxes, and the answer engine names a handful of businesses and leaves the rest unmentioned. Businesses that can afford to optimize for how these systems read and recommend get named. Businesses that can't afford it quietly stop showing up in the answer, with no error message, no warning, and often no idea it happened at all.

Neither future is inevitable. Nothing about the technology itself forces the entrenching outcome. What decides it is a much older and more boring question: who builds the tools, and who they build them for. That question is the entire reason Surmado exists. Small businesses should receive the capabilities of AI without having to become AI companies. That's the belief the company is built on, and the rest of this essay is really just me working out what it requires in practice.

The lesson from electricity

The clearest precedent I know for what's happening isn't a technology story. It's an infrastructure story, and the first time it played out, it took about thirty years.

Electrification did not raise factory productivity when the dynamo showed up. Steam-era factories were built around one enormous engine, with belts and shafts radiating out to every machine on the floor, so bolting an electric motor onto that same layout mostly just got you a slightly better steam engine. The productivity boom arrived decades later, once factories were physically torn apart and rebuilt around electric power: a motor at each machine, a floor plan nobody could have used under the old system at all. That's Paul David's argument in The Dynamo and the Computer (American Economic Review, 1990), and it's one of the more useful papers I know for understanding what's happening to us now.

The same lag shows up in general-purpose technologies broadly, not only electricity. Complementary investment in new processes, retrained people, and restructured data has to happen first, and the payoff stays mostly invisible in the numbers while that work is underway, which is why measured productivity can actually dip before it rises. That's the productivity J-curve, and it's the shape of Brynjolfsson, Rock, and Syverson's work on the subject.

Big companies are doing that complementary work right now. They have the budget to hire people whose entire job is redesigning a workflow around a new model, and the patience to absorb a couple of quarters of dip before the curve turns up. The U.S. Census Bureau's own numbers show the gap opening in real time: 37% of businesses with 250 or more employees report using AI, against under 20% of businesses with four employees or fewer. A small business can't take that path. You can't close for six months to redesign your shop floor around a technology you're still learning to trust. You have customers today.

AI is like electricity. Surmado builds the factory: pre-redesigned, $99 a month.

That line is really the whole business model in one sentence, so I want to be precise about what it means. We aren't selling you electricity. We're selling you the factory that's already built to run on it, so you skip the thirty-year redesign and start with the version that already works.

What AI for small business actually means

AI makes it easier to run your business. It does not run your business. Easier is not the same as automatic, and you stay in charge of what "good" looks like. Nothing we build is designed to change that.

In my research, I call this answerability. A machine can execute work, but someone must still decide what the work is for, what good looks like, whether the result meets that standard, and whether it should enter the world. I set out the full argument in a preprint called The Decision No One Authored, but the plain version is the one that actually matters day to day. Work can be delegated. Responsibility for whether it was done well cannot disappear. A website Surmado writes, a report Surmado generates, a recommendation Surmado surfaces: all of it is still yours to approve, reject, or send back.

This is also why I flinch at the word "replace." We aren't trying to replace the people you already employ: the bookkeeper, the person at the front desk, whoever answers the phone when a customer is upset. We're trying to give the people you already have new capabilities, with guardrails around where those capabilities are allowed to act on their own.

I've watched the alternative up close. A business hires an agency to build a custom AI agent, pays tens of thousands of dollars for it, and gets something that works fine in the demo and then does something strange once it's live: sends the wrong email, quotes the wrong price, promises a delivery date nobody can hit. When it breaks something, no one is clearly answerable for it, because no one signed up to own that failure.

The answer is not less automation. It is framed automation: hold the points where judgment enters force, automate the execution between them, and keep failure owned.

I wrote about this at more length in a preprint called Building Answerable AI. The short version is that you don't have to choose between a human doing everything and an agent doing everything. You choose where a human has to look before something ships, and you let the software move fast everywhere else. That principle binds us too. We publish the standard our own products have to clear before we call them finished, the same standard a customer can hold us to.

Big companies have teams to turn new technology into an advantage. Small businesses usually get another tool, another login, and another job to do. Surmado exists to close that gap. Part of how we close it is deliberately unglamorous: hosting, security, monitoring, an operations team watching for problems at three in the morning. That's infrastructure a large company budgets for as a matter of course, and a small business normally can't justify buying on its own. We buy it once, run it well, and spread it across many businesses instead of asking each one to build its own version from nothing.

The bigger map

None of this is really about one country, one language, or one kind of business. What I actually want is AI cheap enough and accessible enough that a business anywhere can use it productively, in whatever language its owner actually thinks in, without the churn, the risk, or the quiet worker harm that the word "automation" usually smuggles in unexamined. Surmado runs in seven languages today, not as a localization checkbox but because the belief this company is built on doesn't come with an asterisk for geography. If it's true, it has to be true everywhere the technology reaches, or it isn't really true anywhere.

I want to end on the correction I'd make if I could only make one. It would be easy to read an essay like this and conclude that Surmado is the fix: the one mechanism standing between small businesses and getting left behind. That isn't true, and I don't want to leave the impression that it is. A broad AI boom will require thousands of mechanisms that translate frontier capability into usable capacity for ordinary firms. Surmado intends to be one of them.

If you're an owner trying to work out what any of this means for your business, the fuller picture is on AI for Small Business. If you'd rather help build the mechanism than just use it, you can learn about the company.