AI writes code fast. On a big project, that's not the hard part.

AI has changed how fast software gets written. We use it every day, and it makes us dramatically quicker — there's no going back, and we wouldn't want to. But if you're paying to have something serious built, there's a limit worth understanding, because it doesn't show up on day one. It shows up in year two, in your maintenance bill.
And it isn't about whether the code works. AI writes code that works. The limit is something harder: thinking in systems.
AI is brilliant at the thing in front of it
Ask an AI to write a function, fix a bug, or build a screen, and it's genuinely excellent — often faster and cleaner than a person doing the same small task. That's real, and it's not going away.
But a large software project isn't a thousand small tasks in isolation. It's a system that has to hold together — and keep holding together as it grows and changes for years. That's where the gap between a seasoned engineer and an AI shows up.
What the experienced human brings
A good engineer carries something an AI doesn't: a feel for where the project is heading. They've built enough systems to sense which parts will be reused, which will change, and where it's worth spending an hour now to save a hundred hours later. They'll stop and say: "we're going to need this in fifty places — let's build it once, in one place, so the day it changes, we change it once."
AI, left to its own devices, doesn't do that. It solves the problem in front of it — well — and moves on. Ask it to fix the same issue in fifty places, and it will dutifully fix it in fifty places, rather than tell you those fifty copies shouldn't exist.
Why AI defaults to "just fix it here"
It isn't stupidity — it's what it learned from. AI has read a sea of small code examples: tutorials, snippets, "how to do X." Almost all of them solve one problem in one file, beautifully. Very few teach it to step back and design something that has to live for a decade. So its instinct is "solve it here," not "make sure we never have to solve this in a hundred places again." Point it at the bigger picture and it can reason about it — but it rarely volunteers to. Someone has to be steering.
The cost you actually feel: upkeep that grows with the project
Here's the part that reaches the bottom line. When the same logic is copied into fifty corners of a system, a single change becomes fifty changes — fifty chances to miss one, fifty things to re-test. The hard part of software was never building it. It's living with it.
On a small project you'll never notice. On a big one, it's the whole ballgame. If nobody designed for change, the effort to maintain the software starts growing as fast as the software itself. Twice the system, twice the upkeep. That's the line between software that quietly makes you money and software that slowly becomes a money-pit.
The rule that makes custom software worth it
For custom software to pay off over years, one thing has to be true: the effort to change it has to grow slower than the software does. Add ten times the features, and it should cost far less than ten times the effort to keep running. Ten times the customers shouldn't mean ten times the maintenance.
That doesn't happen by accident, and it certainly doesn't happen from "just make it work." It happens by design — shared foundations, one home for each piece of logic, and someone thinking a step ahead of the code. It's the difference between a system that gets cheaper to change as it matures, and one that gets more expensive with every feature you add.
Where the human earns their keep
None of this is an argument against AI. We lean on it hard — it's the fast hands. But the value we bring a client isn't typing the code; it's the judgment around it: deciding what to build once and share, spotting the thing that will be changed a hundred times and making it a single change, and keeping your software cheap to change as it grows. AI accelerates the work. Experienced people keep the whole thing from turning into a maintenance trap.
That's the pairing that actually works on serious projects — AI's speed, with a human who has seen where this road goes.
So if you're having software built — by us or anyone, with AI or without — the question to ask isn't "does it work today?" Almost anything clears that bar now. The better question is: "how much will it cost to change next year?"
That's a conversation we're always happy to have. Tell us what you're trying to build, and we'll be straight with you about how to keep it cheap to own.