I'm hearing the same story more and more often at the moment.
A builder gets on the phone and tells me he had a crack at building his own app over the weekend. It usually starts with something small and genuinely useful. A tool that pulls the weather forecast for the site postcode and flags which days on the program are at risk of rain, so he knows on Sunday night whether Tuesday's pour is on. He typed what he wanted into an AI, it wrote the code, and by Sunday afternoon it was working on his phone.
He's rapt. And honestly, he should be. That's a real thing he made, and two years ago he couldn't have made it at all.
Then comes the question sitting underneath the story. If I can build that in an afternoon, what exactly am I paying a software company for?
That's a fair question and it deserves a straight answer rather than a defensive one.
What the weekend build actually proves
What he built is real. What it isn't is finished.
The weather tool works because it does one thing, for one person, on one phone, with one set of assumptions, and because the person using it is the same person who built it. When it breaks he knows why. When the forecast looks wrong he just ignores it. There is no one else depending on it.
Now change the conditions. Put fifteen people on it. Have it write to your program so the dates actually move. Have it notify the concreter, and then have the concreter's price change because the pour slipped. Have it hold data you are legally required to keep for seven years. Have it keep working when the API it depends on changes its response format without telling anyone, at 4am, while you're asleep, on the week you're chasing a certificate of occupancy.
That's when you find out what software actually is. The demo is the easy part. The hard part is everything that happens after the demo, and it never ends. Data integrity, permissions, backups, uptime, security, migrations, someone leaving the business and taking their login with them, and the thousand small decisions about what happens when two people do the wrong thing at the same time.
AI has made the first 10% almost free. It has not touched the other 90%, and the other 90% is where all the risk lives.
The piecemeal trap
The other thing happening right now is a flood of small software doing single tasks. One app for site photos. Another for variations. Another for timesheets. Another for safety. Each one is cheap, each one demos beautifully, and each one solves a real problem you actually have.
So you buy them. Nine months later you're running eight or nine tools, and none of them talk to each other. Your variation lives in one place, the photo that proves it lives in another, the timesheet that costs it lives in a third, and the client email that approved it lives in Gmail. Your admin person spends her mornings copying numbers between systems, and when the client rings and asks why they're being charged for something, nobody can answer the question in under twenty minutes.
You didn't buy software. You bought yourself a data entry job and a dispute you can't win.
This is the part that gets missed in the excitement. The value in a build business was never in any one of those functions. It's in the thread that runs through all of them, from quote to variation to invoice to handover, where each step carries the evidence from the last one. That thread is the product. You cannot vibe code a thread.
Why every hire we make is a software engineer
We made a deliberate decision early on, and I'll be blunt about it because it's the whole point of this piece.
Every hire we've made at Paperless is a software engineer. Not growth people. Not a sales floor. Engineers.
That's not because engineers are magic. It's because when your customer is trusting you with their contracts, their client communications, their variations and their money, the quality of the engineering is the product. Anyone can produce a screen that looks right. Producing a system that is still correct in three years, after fifty releases, across thousands of jobs, with nobody's data leaking into anybody else's, is a completely different discipline. It's earned, not prompted.
We live and breathe software. It's not a department here, it's the whole company. And we use AI heavily ourselves, every day, because we'd be mugs not to. The difference is what happens after the AI writes the code. It gets reviewed, tested, security scanned, staged and monitored by people who have spent their careers learning what goes wrong. AI made our engineers faster. It did not make them optional.
If anything, the flood of AI-generated code makes real engineering more valuable, not less. When everybody can produce something that looks like software, the only thing worth paying for is software that holds.
We've been here before, every generation
None of this is new. The industry has run this exact panic three or four times in my working life.
When the calculator arrived, did it get rid of the architect? Of course not. It got rid of an afternoon of arithmetic and gave that afternoon back to design. The architect who used it got more done. The profession didn't shrink, it moved up.
When robotic and prefabricated concrete wall systems came in, was that the end of carpenters? Have a look around any site today. The good carpenters didn't disappear, they moved to the work the machine can't do, and plenty of them ended up running the crews that install the panels. The trade changed shape. It didn't die.
And now everyone's watching modular. When you read that a volume builder can put a home up in twenty days using components shipped in from overseas, it feels like the ground is moving. But somebody still has to secure the land, run the approvals, coordinate the trades, manage the client, handle the defects, and carry the risk. The twenty days is the visible part. It is not the job.
Efficiency arrives in every generation. It always looks like a threat on the way in, and it always turns out to be a tool. The people who lose are the ones who ignore it. The people who win are the ones who pick it up and point it at something useful.
What actually decides who's still standing
I've been managing and growing businesses for fifteen years now, through a few cycles that ate a lot of good operators. The formula has never changed once, and AI hasn't changed it either.
The businesses that last are the ones with process. Repeatable method, so the outcome doesn't depend on who happened to be on the job that week. Clarity with clients, so expectations are set before the money is spent rather than argued about after. Evidence, so when there's a dispute you open a file instead of a memory.
That's it. That's the whole thing. Every business I've seen grow fast and stay grown had those three. Every one that blew up had brilliant people and no system.
Software doesn't create that discipline. It can only carry it, or expose the lack of it. A builder with real process and a spreadsheet will beat a builder with no process and forty apps, every time. And a builder with real process, running on a system built by people who take software seriously, will beat both of them and go home earlier.
So build the weather app. Genuinely, have a go, you'll learn something and you might solve a real annoyance. Just be honest with yourself about what it is. It's a tool, not a business system.
The winners in this next stretch won't be the ones who used AI to make an app. They'll be the ones who used the time it gave them back to tighten their process, sharpen their method, and give their clients the clearest experience in their market.
That's been true for every generation of tools. It's still true for this one.