AI writes the code now. So what am I for?
Building software gets cheaper every six months. An honest look at what that does to a one-person automation consultancy, why cheap to write is not the same as safe to run, and where the value actually sits.

Daniel Huthwaite
Founder, 88 Automate
A year ago, connecting a booking system to a payment provider and an accounts package was a week of careful work. Now it's a day, and most of the typing is done by an AI while I check it. Every six months it gets cheaper again. If you run a small consultancy that builds systems, that's an uncomfortable trend to look at honestly. So here's the honest look.
The build is the part that's getting cheaper
Let's not pretend otherwise. Wiring systems together, scaffolding a portal, writing the integration code: this is exactly the work AI does well, and clients increasingly know it. If my income depended on charging a lot for building, I'd be selling the thing that's losing value fastest.
So I don't. Every build I do is quoted in hours and fixed before we start, and the price is deliberately competitive. The build is not where I make my living. It's how I get inside a business.
Cheap to write is not the same as safe to run
Here's the part the "AI builds your app in an afternoon" crowd leaves out. Getting code written is now the quick bit. Getting it to a state where it can take your customers' money and hold their data is not, and it never was the typing that made it slow.
Before anything I build touches a real customer, it goes through the same checks it always did, and AI has made some of them more important, not less:
- Testing against the real thing. Not "it worked on my screen". The checkout is tried with a real card, the invoice is checked in the accounts package, the email is opened in an inbox. This summer a change that looked fine in testing refused every online payment for a client for several hours, because it read the wrong field. Nobody's tool caught it. A person checking the live system did.
- Security, line by line. AI writes plausible code, and plausible is exactly how security holes look. Who can see this record? Does this button check that you're the right customer, or just that you're logged in? Is that API key sitting where a browser can read it? I've found all three in code that was generated quickly and looked fine. Every build gets a review for who can reach what, and secrets never leave the server.
- Checking that it actually went live. The deployment tool says done. Did the change reach the live system? Did the database update actually run? I check, because I've been burned by trusting the green tick.
- Watching it afterwards. Nothing stays working on its own. Health checks on every connection, and a person who notices when one goes quiet.
That's the work that doesn't get cheaper, because it's judgment applied to a real business with real money moving through it. When someone offers to build your system in an afternoon, ask them what happens in the week after.
What the build actually buys
Once a system I built is running inside your business, I can see how it really works. Which enquiries convert. Where jobs stall. Which customers come back. Which month the margin quietly slipped. No outside consultant sees that, and no general-purpose AI tool sees it either, because it isn't in the data anyone hands them.
That's the part that doesn't get cheaper. It gets more valuable the longer I'm there.
The three things AI can't do for you
Know your business. An AI can analyse a dashboard. It doesn't know that your best salesperson is unreliable in January, that the real constraint is staff supply rather than demand, or that the last time you discounted it wrecked your margin. I know because I've been inside the operation for years, and every month I look at the numbers with that knowledge.
Carry lessons between businesses. Every client teaches me something: what converted, where deals stall, which automation actually moved revenue. A system I designed this year for an accounting practice was shaped by a problem I hit at a party company the year before. Your own AI subscription will never have that. Mine gets better with every business I work with.
Be accountable. When something breaks at two in the afternoon, someone has to notice, care, and fix it. A model doesn't answer the phone. I do, and my name is on it.
What this means if you're hiring someone like me
Ask what happens after the build. If the answer is "we hand over and you call us if something goes wrong," you're paying for the part that's getting cheaper and nothing else. The consultancy that's worth paying for in 2026 is the one that stays: watches the system, reads your numbers every month, tells you what to do next, and builds it.
AI writes the code now. That's genuinely good news for you, because it makes the first step cheaper. What you're paying for is the judgment about what's worth building, the context to know it, and someone who's still there next year. That's the job, and I don't think it's going anywhere.
Sound like your business?
Thirty minutes on a call with me. You tell me what's frustrating you, I tell you straight what I'd do.
More notes
How to automate your business processes — the 7 steps that actually stick
Most automation fails because it starts with the tool instead of the process. The 7-step method I use with every client: find the repeat, put a number on the leak, fix the process, then automate the smallest piece first.
Five apps, and no screen showed the whole client
A bookkeeping practice with 84 clients and five good tools that didn't talk. Why we diagnosed before building, and why we didn't migrate a single record.