Research shows that three out of four small businesses are experimenting with AI — but fewer than one in ten have actually gotten it to work. The most common reason they're stuck? "We don't know where to start."
That gap is exactly where the tools move in. Duda says it makes building a website as easy as making a slideshow. N8N says it can automate everything you repeat. The boss has a problem, the tool says it can solve it, and the money goes out.
But from what I've observed, the problem isn't about picking the right or wrong tool. It's more fundamental than that. Part of what you're paying every month is subsidizing the vendor's business logic — not your own.
Their Success and Yours — Are They Even the Same Thing?
Duda succeeds when your website goes live and the monthly fee keeps coming in.
You succeed when a customer picks up the phone or fills out an inquiry form.
An N8N consultant succeeds when the workflow has more nodes, more integrations, more complexity — the more you depend on them, the better.
You succeed when that thing you do manually every week simply disappears.
These two definitions don't line up. And you won't notice it at the point of purchase — usually it's months later, fees still going out and results never materializing, that something starts to feel off.
Klarna isn't a small business, but they ran into the exact same dynamic. They cut Salesforce and Workday, saving $40 million a year. The CEO's explanation was blunt: Salesforce's success is you continuing to use it, not your business growing.
That's not saying those tools are bad. Salesforce is genuinely powerful. Duda's interface is reasonably intuitive. The issue is that their business models were never designed to let you "use them until you no longer need them." Renewal rates, engagement stickiness, upsell paths — those are the numbers they're actually optimizing for, not your revenue.
You don't know this because nobody tells you during the sales process. By the time you figure it out, you've usually already paid for a full year.
Small businesses don't have a finance department running these calculations. But monthly fees and headcount are equally real costs.
Whatever Happened to the Tools That Were Going to "Lower the Barrier"?
Benedict Evans, former a16z partner, titled his annual report last year AI Eats the World. His argument: software ate traditional industries; now AI is eating software itself. The middlemen who survive by sitting between you and the answer are disappearing, layer by layer.
Chegg is the clearest example. It was America's largest online homework-answer platform. The business model was simple: sit between students and answers, charge monthly.
November 2022: ChatGPT launches.
Five months later, Chegg's CEO mentioned ChatGPT hurting the company on an earnings call for the first time. The stock dropped 48% that day. Then kept falling. Market cap went from $14.7 billion to $110 million — a 99% collapse — in three years.
They tried to save themselves. They built their own AI version using GPT-4, called CheggMate. It didn't work. Students had no reason to choose the packaged version over the original. The lesson is brutal: you can't fight a stronger thing by wrapping it in packaging.
Jasper AI's story is even more direct. It was itself an AI tool, built specifically to write marketing copy, valued at $1.5 billion in 2022. After ChatGPT appeared, revenue was cut in half within a year. Founders departed, the CEO changed.
Even if your tool is built on AI, if you're sitting between the user and the answer, you still get eaten.
Duda's problem sits at the same layer, but closer to you. Its pitch is "building a website as easy as making a slide deck." But slide decks themselves are nearly gone as a manual task now — tools like Gamma can generate a complete presentation from a few sentences of description. The very thing Duda used as its analogy is itself close to being replaced.
The more practical question: you can tell Claude directly what kind of page you want, and what comes out isn't slower than learning Duda's interface — and the result isn't necessarily worse. And Duda never solved the hard part anyway: how to configure your domain, how to point your DNS. You still have to figure those out yourself, or find someone else to handle them. AI can walk you through every step.
The barrier Duda was going to lower was never as high as they claimed.
One more thing worth knowing. Duda's pricing page states it plainly: data export is only available on the highest-tier Agency plan. The articles, pages, and images you put in — if you ever want to switch platforms, you either pay more to upgrade, or you rebuild from scratch.
Easy in, expensive out. That's not an accident. That's by design.
So What's Actually Wrong with N8N?
Conceptually, nothing. The need to automate workflows is real, and N8N's feature set is genuinely rich.
The problem is maintenance.
A workflow takes time to build. Once it's running, you enter "observation mode." Then one morning you get a notification: the workflow broke. You go in and find that some external service quietly changed its API format — that node can't read the data anymore, and everything downstream is stuck. You find someone to fix it, it runs for a few days, then something else goes wrong.
This isn't unique to N8N. N8N's own community forum has a thread titled "When N8N is NOT the Right Choice for AI Automation." Not an outside critic — heavy users themselves saying it.
A workflow can only handle perfectly the 80% of situations it was designed for. The remaining 20% — wrong format, service timeout, missing field — requires human judgment every single time. Over time, the maintenance cost exceeds the time saved. And every person who can fix it is now carrying an invisible maintenance burden in their head.
AI agents handle this differently. You say what result you want; the agent finds a path to get there. When something goes wrong, it tries to fix itself first — and only tells you what happened if it can't. Fault tolerance is built in, not dependent on you having pre-drawn every possible exception into the diagram upfront.
This gap isn't about feature strength. It's about a different way of thinking. N8N's logic: you pre-define every step, it executes. Agent logic: you tell it the goal, it figures out how. The former requires you to have the problem completely thought through before you start. The latter lets you think through the problem as you go.
For a small business with no engineers, that gap is very real.
What you're building isn't expertise in some tool's node logic. It's the ability to clearly articulate the problem you're trying to solve — and that skill travels with you no matter what tool you switch to next.
So What Should You Actually Do?
I know some readers will get here and say: "Fine, I get it — but I don't have time to figure all this out myself."
That's exactly the line those tools love to hear. Because "the owner is too busy" becomes the justification for "the owner needs me."
Finding someone to execute is fine — nothing wrong with that. The question is: who are you handing your budget to?
Shopify CEO Tobi Lütke put it plainly in an internal memo last year:
"Using AI effectively is now a fundamental expectation of everyone at Shopify. Before asking for more headcount and resources, teams must demonstrate that they cannot get what they want done using AI."
He didn't say "have IT evaluate AI tools." He said: take your own work and run it against AI first. Buying tools comes later — it's not the first step.
Wharton professor Ethan Mollick has studied AI's impact on work for over three years. His conclusion is a single sentence:
"Use frontier models to figure out what they do. Use these models a lot."
Not taking a course. Not going through someone else's packaged interface. Just having employees take their real daily work and run it directly against the model. MIT research found this approach reduces task completion time by an average of 37%. BCG's study of 800 consultants found the same.
"But isn't AI hard to learn?"
NVIDIA CEO Jensen Huang said at Davos this year:
"AI is the easiest software to use in the history of software."
You don't need to know how to code. You don't need to set up a workflow. You don't need to read documentation. You just need to explain your problem clearly — and "explaining your problem clearly" is something you're already doing every day.
Practically: take the monthly fees you're paying for packaged tools and swap them for Claude or ChatGPT subscriptions for your employees — one account each. Today, use it to organize client data. Tomorrow, use it to draft a follow-up email. The day after, have it look at last week's sales numbers and flag what seems off. No need to design scenarios — just throw whatever you have at it.
Over time, each employee will develop their own feel for it: what can be handed off to AI directly, and what needs their own judgment. That judgment can't be purchased. It only comes from use. And it belongs to that person — not to a tool subscription. When you switch platforms someday, they take it with them.
The next time someone comes to sell you an AI tool, ask two things:
"Can I get the same result by using ChatGPT or Claude directly? If not, what specifically can't it do?"
Chegg couldn't answer that. Jasper couldn't either. If the answer is "our interface is nicer" or "we've integrated everything for you," that's packaging, not capability. A tool that's genuinely worth paying for should be able to name one concrete thing the underlying model can't do — but it can.
"After using it, which specific action goes from minutes to seconds? Can you give me a real example?"
N8N takes time to set up, time to maintain, time to fix when it breaks. A lot of tools' "time savings" just shuffle time from one place to another — it doesn't actually disappear. A tool that can say "this step used to take 20 minutes, now it takes 30 seconds" is the real thing.
If they answer both questions clearly, keep talking. If they can't, your time is already saved.
The last post was about AI amplifying what you already have. This one is the other side: tools that claim to lower the barrier sometimes just add one more toll gate between you and the answer.
Walk straight through.
Related reading
If you want to understand the flip side — what AI can amplify when you use it directly — the previous post covers that: Getting Started with AI: Amplify What You Already Do Well.
Tell me where you're stuck
I'll get that process running with AI and make sure your team can keep it running on their own.
Talk to Barry