TLDR: Small business AI use fell from 42% to 28% in a single year, which means millions of owners tried the tools and quietly walked away. Most of them were following directions that stop early. This piece gives you three conditions that mean you should hire someone instead of building it yourself, along with the questions to answer before you decide either way. For two essays a week on building systems from expertise, subscribe to Unpromptable.
So, you know that I just recently moved out of my hometown. What I didn’t tell you is that in the place I moved into, the toilets didn’t work. Two bathrooms, both with non-functioning toilets.
I’ll spare you the gruesome details, but I’ll leave you with this: I consider myself a pretty handy guy (as many people probably do). I know how to figure stuff out, know my way around tools, all that. But I took one long look at that system and went, “I need a plumber” (also, I thought, “I need to argue with my landlord,” but that’s a different story).
There are problems we know how to fix. There are problems we’ll eventually figure out. And there are problems that we can figure out, but really don’t have to.
Unfortunately, this is how we’re talking about AI.
New here? I’m James. I help founders and creators build AI systems that make them impossible to replace. My mission is to help you use AI to become irreplaceable at what you do by turning your expertise into systems no one else can replicate. Here’s an example of my work, turning a therapist’s 20+ years of experience into an AI-powered diagnostic.
AI has taught everyone that they can do everything themselves, or rather, that they can do everything by asking AI themselves.
This has some truth to it. AI multiplies what you can do by a factor of 100x. Writing for non-writers, coding for non-coders, analytics for the non-analytical.
It also has limits, and this is what AI influencers fail to acknowledge.
Your capability is only limited by what you know to prompt in the first place. If you don’t know that something is possible, it won’t be possible. If you don’t know that something HAS to be done, it won’t be done.
Now apply that to an AI project inside your own practice.
You’re a coach, a consultant, or any number of credentialed expert, thirty years of practice behind you. You decide to build an AI assistant that talks to leads the way you would, because you know AI can do it and you’ve watched other people pull it off.
Three months in, one to two hours a day, and prompting turned out to be the smallest part of it. The thing answers questions about your own framework and gets it wrong in ways that make your chest tighten. You’ve realized that deploying it introduces practical and legal risks that you haven’t even heard about before then.
It’s not ready at all. And worse, those hours came out of the practice that pays you.
You’re becoming a real builder, for sure, but it’s costing you time, energy, and money. When all you wanted was the product.
Don’t get me wrong, I’m not saying you shouldn’t ever tinker with AI. You should definitely build something cool with it, some time.
But with how powerful AI has become, and with how much influencers hype it, many founders believe it’s the answer to their business problems. In many cases, it’s not. It can be, but you need to know when, and where.
Here are three rough criteria to decide this by:
When experiments light you up. If you have ever lost a Saturday to a workflow and looked up at 9pm happy about it, you’re the person these courses were built for. That feeling alone could be the payment for a long stretch of frustrating work. I took a version of that road myself, and it gave me a career.
When you have time and energy beyond your practice. The real cost was never the course fee. It’s the hours after client work, when the judgment your clients pay for sits idle while you debug.
When revenue doesn’t depend on it. Learning to build pays back in seasons. If this quarter’s number is the thing keeping you up at night, building an AI workflow yourself will likely just add to your problems. It’s the wrong road this quarter. It might be the right one next year.
AI educators are not lying to anyone. They’re just handing out maps to a journey suited for a specific kind of people, ones with disposable time, energy, and income. As founders, business owners, and creatives, you might have all these three — and that’s perfect.
But in many more cases, experts don’t really want to be doing more than what’s already making them money.
Something happened to small business AI adoption between 2024 and 2025 that nobody selling courses talks about. It went backward. A survey of 1,500 small business owners found usage fell from 42% to 28% in one year. Millions tried the tools, and enough of them quietly walked away to drag the national number down.
The walking away is expensive. When a project dies at a small business, the sunk cost lands between $25,000 and $100,000 in time and abandoned work.
Bigger budgets tell the same story. MIT’s NANDA project found that 95% of generative AI pilots inside large companies produced zero measurable return, meaning nobody could point to a number that moved after the money was spent. The same research showed systems built by outside specialists reaching deployment at about twice the rate of in-house builds. And S&P Global found enterprises abandoning most of their AI initiatives jumped from 17% to 42% in a year.
Teams with engineers on payroll and dedicated budgets stall when they realize the complexity of what they’re trying to accomplish, and how AI prompting is just a tiny step in this process.
You’re working solo, after client hours, with none of all that.
Experiment all you want. Prompting, testing tools, seeing what a chatbot can do with your intake questions. That work is cheap and it teaches you things.
Past a certain point, though, you’re not experimenting anymore. You’re running a project. And a project needs someone who has finished one before.
That someone is a builder, not another course. A course sells you capability. A builder sells you the thing itself. Both are real products and both have their place. The question is which one you’re short on. If you’ve already bought capability twice and your system still doesn’t exist, you were shopping in the wrong aisle.
So where’s the line? Three criteria draw it.
Some AI work fails privately. Your summarizing workflow breaks and you lose an afternoon.
Some fails in front of the people who pay you. For example, I built an assessment for a therapist that her audience takes before their first session. It’s the first thing they experience of her practice, and it runs on twenty years of clinical judgment. That build had to be right, because a bad version wouldn’t just be useless. It would have told a hundred strangers something untrue about who she is.
Private failure costs you time. Public failure costs you the thing you spent twenty years earning. Here are some questions to ask about the build:
Will a client, a lead, or a reader touch this directly?
If it produced something embarrassing tomorrow, who would see it?
Does it speak in your voice, or answer questions people are trusting you to answer?
Pain is a bad signal on its own. Pain at 11pm is what makes you buy material you will abandon once your practice consumes your time.
The better question is what another six months of this costs you. Not how much it hurts today, but the hours, the dropped follow-ups, the leads that went cold while you meant to get to them.
Run the calculations here. It’s usually worse than you think and it makes the decision for you.
How many hours went into this last month?
Multiply by twelve. Would you pay that to make it stop?
Have you already paid for a tool that was supposed to fix it?
You could learn retrieval systems, embeddings, evaluation, and the rest of it. You can learn how AI works for marketing.
It would take months, and at the end you’d hold a skill that has nothing to do with the work your clients pay for.
When the knowledge is far enough away, you can’t tell AI what you need either. Your capability is limited by what you know to prompt in the first place, so a gap in domain knowledge is a gap in your prompts. AI will answer the question you asked. It won’t tell you that you asked the wrong one, and it won’t mention the thing you didn’t know to ask about.
That’s how three months disappear. The tools worked fine. Nobody in the room knew what was missing.
Would learning this build a skill you’d use again after this project?
When the output looks wrong, would you know why?
Could you tell a good version of this from a mediocre one?
Hire a builder and you depend on the builder. I hear this one a lot, and it deserves a straight answer.
The plumber leaves. The pipes stay in your walls, the toilets stay fixed, carrying your water.
A system built from your expertise works the same way. Your frameworks power it and your judgment shapes every answer it gives, which means it belongs to you the way your pipes belong to you and no plumber ever will.
Compare that to the dependency nobody talks about: an expert running client work through generic tools that hold none of their thinking, where every output could have come from any stranger with the same subscription.
There is a word for that kind of output: promptable.
It carries nothing of you, and clients feel that even when they can’t name it. A system built from your archive and your judgment makes work no stranger can reproduce.
That word sits on the masthead of this newsletter because it’s the point of hiring a builder: you walk away holding a moat, and the moat is made of you.
Done right, hiring is how you end up owning more.
I never learned plumbing. I made a call, argued with my landlord (although argued is a strong word, he had no leg to stand on, knew it, and paid for everything but the hassle), and by the end of it the problem was gone.
I got two working bathrooms and my attention back. The plumbing knowledge stayed with the plumber, where it belongs.
Building your own AI system or product is exciting, like a roller coaster ride, filled with its own ups and downs. But having a system built for you is smoother, quieter. The follow-ups happen without you. The intake runs itself. The knowledge that lived only in your head now works hours they don’t.
That’s the outcome worth wanting. Your own expertise, running without you in the room.
Just because you can doesn’t mean you should. You can build it. Give it enough Saturdays and you’d likely finish. The question is whether the version of you who finished it is better off than the version who spent those Saturdays on the practice that pays.
A system built from your expertise belongs to you either way. You don’t have to be the builder to own it.
You don’t have to be swimming, or drowning, to ride the AI wave. You can charter a boat.
If this way of thinking is useful to you, subscribe. Two essays a week on building systems from expertise, real numbers and sideways builds included.
PS. The AI Readiness Diagnostic takes 5 minutes and tells you whether an AI project makes sense for where you are, including when the answer is no.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.