TL;DR
AI assembles your client’s tool in minutes. The screen shows up, the buttons respond in the demo, and it looks like it’s done. It isn’t. What AI delivers is the part you can see. What it doesn’t deliver is what holds the tool up when real people use it: the rule that can’t miscalculate, the data of everyone who signed up, and what happens when a thousand people hit it at the same minute. The agency ships it thinking they solved the problem. The problem shows up later, in front of the end client, with your brand on the screen. The cheap part isn’t the tool. It’s the bill that comes after.
AI makes the screen. The problem was never the screen.
Today anyone can generate a tool with AI in an afternoon. You describe what you want, it hands back a calculator, a form, an assistant that answers on its own. Looks like magic, and on the visual side, it is. The trap lives there: the screen was always the easy part.
I’ve worked in software for over ten years, much of it on systems that handle money and can’t miscalculate. The lesson that repeats most is this: what takes real work was never drawing the screen. It was making the thing behave the same for the first user and the thousandth, on a bad phone and a bad connection, on a regular day and on campaign day. AI is excellent at the first part. It has no way to guess the rest, because the rest depends on your business, not the visual.
“But it worked when I tested it.”
It worked with you, alone, on a good connection, with nobody trying to break it. That isn’t the test. That’s the rehearsal.
What AI doesn’t deliver (and nobody warns you about)
The AI-generated tool delivers the facade and leaves out the foundation. Five things decide whether it holds up with the end client, and none of them show up in the demo:
- The rule that can’t miscalculate. The calculator shows the right number in the demo and rounds wrong on an edge case. Nobody notices until your client’s client closes a deal at the wrong price. Then the problem has your agency’s name on it, not the AI’s.
- The security of the data users enter. The minute the tool asks for email, phone, or any data, it became a vault. A tool generated in a rush tends to leave the key in the lock: exposed data, open access, passwords in the wrong place. You only find out when it leaks, and by then it’s too late.
- What happens when the campaign works. Scaling: holding up under many people using it at the same time without going down. The tool that responds fast for ten people can freeze for a thousand. And a thousand people is exactly what you wanted: the campaign is working. Success knocks over what was built to impress instead of endure.
- The maintenance the day after. Code generated in a hurry is easy to start and hard to continue. When the client asks for a simple change, nobody understands what the AI wrote, and the “quick fix” becomes a rewrite from scratch. The cheap option comes back as rework.
- The accountability when it breaks. AI doesn’t answer the phone. The freelancer disappeared. When the tool goes down on a Monday morning, one person is left to explain it to the client: you. The tab for the failure is always charged to whoever signed the delivery.
Notice the pattern. None of this shows up on the screen. All of it shows up on the invoice.
“AI has evolved, it handles everything now”
“The models got better. Today AI already takes care of security, scaling, all of it.”
Partly true. AI writes better code today than it did last year. But writing good code isn’t the same as making the right decisions for your specific case. AI doesn’t know how many people will use the tool at launch, doesn’t know which data your client is legally required to protect, doesn’t know the campaign goes live on the 15th. Those decisions aren’t in the code. They’re in the business. And that’s exactly where the tool either holds or breaks.
AI is excellent in the hands of someone who knows what to ask and what to verify. On its own, it delivers what you asked for, not what you needed. The difference between the two is the loss.
How to deliver the tool without falling into this trap
The way out isn’t to throw AI out. It’s to not trust it alone. AI accelerates whoever knows what they’re doing and hides the hole from whoever doesn’t. What closes the hole isn’t a hero on call: it’s a process that reviews what AI produces before it reaches the client. A pipeline that estimates the locked price first, builds, validates, and only delivers what passed, with you approving what works, is what turns “looks done” into “is done.” For the agency, this comes down to three practical rules:
- Whoever delivers has to be whoever answers when it breaks. A tool doesn’t end on delivery day. It lives as long as the client uses it. If the person who built it disappears the next day, you bought a time bomb with your brand on it. Close with someone who stands behind what they shipped, not someone who vanishes.
- Scope and timeline agreed before the first line of code. No open budget that starts at X and triples. You need to know the bill before you promise anything to your client. I wrote about how to price this without fooling yourself.
- Your brand up front, the engineering invisible. That’s the white-label model: the technical partner delivers under your brand and stays invisible to your client. You keep the account and the relationship. The scary part, making the thing hold up without breaking, leaves your desk.
The point isn’t to spend more. It’s to stop paying twice. The first time when you find the cheap option. The second time when it comes back, with interest in reputation. If your team froze before even getting to this point, I cover the beginning of that story in a client asked for a tool and your agency can’t build it. And if what landed in your hands is an already-half-built app that needs to become real, the diagnostic roadmap is in I inherited a vibe-coded MVP to scale.
Looks done isn’t done
AI changed how long it takes to make the screen. It changed nothing about what holds the tool up after it goes live. Security, correct business logic, holding up under real people: that’s still work, still depends on someone who knows what they’re doing, and is still what separates the tool that sells from the tool that embarrasses.
Your client isn’t buying a pretty screen. They’re buying the peace of mind that it works when they need it most, in front of the people they most want to impress. Looks done is what AI delivers for free. Being done is another thing entirely, and that’s exactly where your name lives.
FAQ
Can I use AI to build the client’s tool?
Yes, and it’s even a good idea: AI speeds up a lot. The risk isn’t using AI, it’s shipping what it produces without anyone who understands it reviewing it. AI builds the part you can see. Security, business logic and stability are guaranteed by people who know what to verify. Use AI as an accelerator, not as the one responsible for the delivery.
How do I know if the tool will hold up with the end client?
Ask about what doesn’t show on the screen. What happens if a thousand people hit it at once? Where is the data from registered users stored? Who fixes it if it goes down Monday morning? If the answer is “it worked in the test,” you have a rehearsal, not a tool. Holding up with the end client is exactly what you can’t see in the demo.
Is a tool built with AI insecure?
Not because it was built with AI. It’s insecure when nobody handled the security part, and tools generated in a rush tend to skip that part because it doesn’t show on screen. The moment a tool stores someone’s data, security stops being a detail and becomes an obligation, including a legal one. That requires someone who knows what they’re doing, not just a good prompt.
What’s the difference from hiring a cheap freelancer?
Same trap, different face. The cheap freelancer and AI on its own deliver the facade and disappear when the problem hits. The cost of cheap is never in the initial price; it’s in the rework, the blown deadline, and the tool going down in front of the client. What you want is a delivery that doesn’t come back, not the cheapest one on the spreadsheet.
How much does it cost to do it right?
Less than it seems, and a lot less than it costs to redo. The real expensive thing is the delayed project, what came out wrong and needs to be rebuilt, and the client who stopped trusting you. With fixed scope, you know the bill before you sign and don’t get a surprise in the middle. Predictability is what you’re buying, not hours of development.