Table of Contents
Something shifted in the last year. Building your own tools, your own landing pages, even your own software, used to mean hiring a developer or sitting down to learn to code. Now you describe what you want, and AI builds it while you make a coffee.
I had a client tell me recently that if he were choosing again, he wouldn’t sign up for the expensive system he’s tied into. He’d just build his own. And I understood exactly why he felt that way.
I want to be clear about one thing before we go further. I use AI constantly. For content and for tech audits for instance. This isn’t a warning to stay away from it.
It’s about one decision. AI can make it look like you can build almost anything, but looking like it and doing it properly are two different things. So “can I?” isn’t the useful question. What’s actually worth building is, and how you’d know if there were mistakes. Building was never the hardest part. Knowing what to build is.
Where AI genuinely helps
Let me start with the yes, let’s use AI if you want, because there’s a lot of it.
AI is brilliant for building inside tools that already do the hard work for you. One client built a whole set of landing pages in Kit with AI, and they came together fast. It’s great for making sense of data you already have by spotting a pattern in a spreadsheet you’d have squinted at for an hour. It’ll give you a solid first draft of an audit, a structure to react to.
Notice what those have in common. The tool underneath is already sound. Kit handles the sending and the security. Your spreadsheet is your spreadsheet. AI is doing the fiddly part on top of something that already works. This is the same spirit as using AI to create content without losing your voice: let it help, stay in charge.
And it goes well beyond building things. I use it to get unstuck, when something’s off and I can’t yet name why, talking it through helps me see what I was circling. It’ll turn a rambling voice note into clean notes, or a long call into a summary I can actually use. Ask it to explain what a setting does before you change it, and you go in understanding rather than guessing. It’s a willing first drafter too, an outline, a sales page skeleton, a clunky paragraph reworked until it reads better. It is amazing to help me create SOPs for clients. None of that is finished work. It’s a starting point you shape into yours, with the result right there in front of you the whole time.
The line that matters: cheap and visible, or expensive and hidden
It’s important to think about when to use it and at what stage. There’s a simple way to decide where and when to let AI loose. And it has to do with the mistakes it can make.
Some mistakes are cheap and visible. A landing page comes out wonky, you look at it, you see it, you fix it in five minutes. I assume you were always going to check it, and it is easy to check because the mistake is right there on the screen. That means the fix is easy too.
Most everyday uses fall in this category. A draft email you’ll read before it sends. A social graphic you’ll look at. An outline you’ll rework. If AI gets it wrong, you notice, because the wrongness is staring back at you.
Other mistakes are expensive and hidden. Let’s say you create a tool that stores private information of people. What if that database quietly stores information in a way that isn’t secure? Nothing looks wrong. It works. You’d never know until the day someone else does. And it isn’t only about data. An automation that quietly emails the wrong group, or sends the same thing twice. A checkout that works but gets the VAT wrong. A setting that lets the wrong people see something they shouldn’t. All of it runs happily and looks fine, while the problem builds in the background.
That’s what makes hidden mistakes the expensive ones. Not that they’re bigger, but that you don’t know to look. Nobody’s checking, because nothing is asking to be checked, so the mistake keeps going, and by the time it surfaces the damage is already done.
And notice this isn’t about how big or clever the thing is. A slick landing page is still cheap and visible. A dull little database quietly holding customer details is expensive and hidden. The question was never how impressive the build is. It’s whether you’d catch it when it went wrong.
That’s the line. Use AI freely where a mistake would be cheap and you’d spot it. Be careful where a mistake would be expensive and you’d never see it coming.
The moment to stop: anything holding people's data
There’s one threshold where I’d stop and think hard. The moment something has a login, you’re holding people’s information, and that changes everything.
Someone I know of was building their own client system. The kind where clients log in, and their details live: personal information, financial information, all of it. Building it themselves, with AI, on their own. My first response was, please get someone to check that it’s secure. Because the one thing I know is that these apps are so easy to hack right now.Â
If you’re a hacker, it must feel like heaven.
I’m not saying that for effect. Security researchers went looking recently. One team examined thousands of apps built with these AI tools and found around 5.000 with almost no security in place at all, and roughly 40% leaking sensitive data. Medical records. Financial details. Private messages. Another analysis found more than 400 exploitable weaknesses across AI-built apps, the most common being that any user could reach other users’ information.
And the part that should make you pause is this. The tools that build these apps market themselves as needing almost no technical skill. Then, when something leaks, their answer is that how the app was set up is the creator’s responsibility. Yours.
It doesn’t only happen with big custom builds, either. It’s the same instinct when someone says, let me just pop my email list into AI to tidy it up first. Lovely, so you’re about to hand thousands of people’s email addresses to a chatbot. Maybe don’t. The doorway is smaller, but it’s the same room.
If it holds someone’s personal data, that’s the moment to slow down and bring in a person who knows what they’re looking at. A proven platform carries the security, the backups and the compliance for you. Build it yourself and all of that quietly becomes your job, whether you realise it or not.
It’s hard knowing what to do by yourself and when to bring in AI or another person.
That is why I’m here. Let me help you!
Knowing what to build is the hard part
Even when AI is the right tool, there’s a catch that costs more than people expect.
Let’s go back to those landing pages. They were built, and then I was asked to check them. Which sounds great, but actually AI had built the wrong kind of pages. My time went on auditing what AI had made, working out what was useful and what was wrong, before I could even get to the work that still needed doing.
If the question had come first, “are these the right pages to build, and is this the right way to build them?”, I could have answered that in far less time, and then the building would have gone in the right direction from the start. Same expert hour. One version moves you forward. The other just unpicks what you already made.
That’s the real cost hiding inside “just build it with AI.” Not only the risk of a leak, but the risk of building the wrong thing quickly, and then paying someone to reverse it. Fast in the wrong direction is still the wrong direction.
AI is confidently wrong. It hands you the mistake looking finished, with no flag on it. And believe me, AI gets it wrong a lot too. The number of times a day where I go ‘That is wrong’ ‘That doesn’t make sense’ ‘Actually you got it backwards’ ‘That setting doesn’t exist’…
You also have to be specific with it every time. Experienced developers use AI a lot these days. Which can help them produce faster. But they can look at what it produced and go, that’s not right, there’s a problem in this bit. They’re still doing the checking too.
If you don’t know what you’re checking for, that’s exactly where it falls apart. And you can’t be specific about something you don’t understand well enough to describe.
So how do you decide, and where to get help
So when is it great to use AI, and when is it better to let someone else take care of it? Three questions cover most of it.
Do you actually know what you want it to do? And would you be able to tell if it goes off the rails? This is the question people skip, and it’s the one the whole thing hinges on. Not a vague sense of it, the specific job you need it to do. The structure you actually need, not whatever it guessed. It’s a lot easier to create something if you know what you need.
Is there already a tool built to do this? Most of the time there is, and it’s already secure, backed up and looked after for you. Building your own version of something that already exists is usually the slow, expensive way to end up with a worse one. And if you genuinely can’t find a tool and still think it’s worth building, test the idea cheaply first. A simple quiz or a rough mock-up will tell you whether it’s worth doing long before you commit real money to it.
Does it hold anyone’s personal data, or sit behind a login? If it does, that’s the moment to slow right down. This is the expensive and hidden end of the line, and it’s rarely worth carrying yourself.
If the answer to the first question is no, it might be better to hire some help. To help you decide if it’s worth building, and to make sure the thing that does get built is sound. This is really about knowing when to do it yourself and when to hire help. If that’s where you are, a block of Tech Hours gets you hands-on help from someone who knows what to look for, so you’re not left hoping it’s fine.
The point isn't to fear AI
None of this is a reason to swear off AI. I’d be a hypocrite, I use it every day and it’s made a lot of my work better. I’ve written before about what AI genuinely can and can’t do for a small business.
So let AI build. Just stay clear on what you actually want, and keep an eye on whether it’s right. The tech does the building. You do the thinking.
