AI coding tools genuinely write working code now. Somebody with no background can describe what they want and get something that runs.
They finish some jobs completely. Others they leave half done, and the half that's missing is the half a business depends on.
What they're genuinely good at
These tools are better than most people outside the industry realize.
- Getting to something that runs. A working screen with real buttons, in an evening, from a description.
- Small scripts. Renaming two thousand files, pulling numbers out of a folder of PDFs, reformatting a spreadsheet. A job that used to take an afternoon takes a few minutes.
- A tool only you use. If it lives on your machine and nobody else depends on it, none of the usual warnings apply. Go and build it.
- Explaining code you already have. Paste in something you inherited and ask what it does. This is genuinely excellent.
If that's the shape of what you need, stop reading and build it. You don't need us, and we'd rather say so.
What's still left afterwards
The gap isn't the code. It's everything the code sits inside, and none of it is visible while you're building.
- It has to run when your laptop is shut. Hosting, a domain, and something that restarts it at 3am when it stops.
- Other people have to sign in. And see their own records and not each other's. Accounts and permissions are their own project.
- The data has to survive a mistake. Backups, and a way to undo. The first time somebody deletes the wrong row, this stops being theoretical.
- It has to cope with someone typing something odd. A date in the wrong format, a photo instead of a PDF, two people saving at once.
- Somebody has to fix it in a year. Possibly not you, and possibly when you're on holiday.
- It has to hold customer records lawfully. Where the data lives, who can reach it, and what happens when somebody asks for theirs back.
Every one of those is unglamorous, and none of them are optional once other people depend on it. Getting to "it works on my laptop" is genuinely a couple of evenings. The rest is the job.
The problem you can't see from inside it
You can't review code you couldn't have written.
AI writes code that looks right. Most of the time it is right. The trouble is the case where it isn't, because it comes out with exactly the same confidence as the case where it is, and nothing on your screen distinguishes them.
An experienced person catches those by knowing what usually goes wrong. Without that, the bug ships and you find out from a customer, or from your accountant, or from a number that's been wrong for four months.
This is the real cost and it's the one nobody prices, because it arrives later, and nothing points back to the cause.
Which side of the line you're on
Build it yourself when it's small, it's yours, and a mistake costs you an afternoon.
Get help when any one of these is true, because each one turns a mistake into somebody else's problem.
- Money moves through it. Invoices, payments, payroll, pricing.
- It holds customer records. Names, addresses, health information, anything you'd have to write a letter about.
- Other people depend on it to do their job. If work stops when it stops, it's infrastructure.
- You'd be the only person who understands it. That's fine until you want a holiday, and it's a serious problem when you want to sell the business.
What it costs either way
Building it yourself isn't free, it's just paid for differently. A coding subscription is $20 to $40 US a month, and the real price is your evenings, for longer than you expect, on something that isn't the work you're good at.
Builds here start at $5,000, quoted in full before anything is written, with the code and the data yours at the end. Larger or more custom projects are quoted separately.
There's a middle route worth knowing about. Build the rough version yourself, use it for a month, and find out what it actually needs to do. Then hand that over as the specification. That's a genuinely good way to spend an evening, and it makes the build cheaper because the hardest question is already answered.