I’ve spent 35 years building software. Twelve of those years were inside a large enterprise, building and managing a bill review rules engine used by insurance carriers across the country. I’ve lived through client-server, the web, the cloud, and the SaaS gold rush. I’ve written proposals for custom software projects with six-figure price tags — and I’ve watched business owners wince at those numbers and settle for off-the-shelf packages that almost fit.
So believe me when I tell you: something just changed, and most business owners haven’t noticed yet.
The old bargain: pay a fortune, or bend your business
For decades, custom software followed one economic law: it was expensive because translation was expensive.
You, the business owner, understood your business perfectly. But you couldn’t code. So you hired people who could code but didn’t understand your business. Then you spent months — and hundreds of thousands of dollars — translating your expertise into requirements documents, watching developers build the wrong thing, correcting course, and paying for every iteration.
Most owners rationally gave up on custom and bought packaged software instead. The package cost less, but it came with a hidden tax: you bent your business to fit the software. The vendor built for the median customer, and you’re not the median customer. You know exactly which screen in your current system makes your team groan. You know the workflow your industry-standard platform handles in eleven clicks that should take two. You know the report you’ve wanted for years that the vendor keeps saying is “on the roadmap.”
That was the bargain: pay $100,000+ for software that fits, or pay $500 a month forever for software that almost fits.
What changed
AI coding tools — Claude Code is the one I use every day — collapsed the translation layer.
Here’s what that means in plain terms. Claude Code is a tool where you describe what you want in ordinary language, and it plans, writes, tests, and fixes the software. Not autocomplete. Not a code snippet generator. An agent that can take “I need a system that imports our vendor invoices, flags anything that doesn’t match the purchase order, and routes exceptions to Sandy for review” and actually build it.
The numbers behind this shift are staggering. This isn’t a fringe experiment — AI-assisted development has become the fastest-adopted developer tooling in the history of the industry, and a meaningful percentage of all new code being published today is written by these agents. The tooling is maturing monthly.
But the statistic that matters isn’t about developers. It’s about who no longer needs one.
The expert was always the bottleneck
Here’s what three decades of custom software projects taught me: the hardest part was never the code. It was getting the business knowledge into the code.
The best system I ever worked on succeeded because the people configuring it deeply understood claims, billing rules, and jurisdiction-specific regulations. The worst projects I saw failed because brilliant developers built exactly what the requirements document said — and the requirements document was wrong, because the person who wrote it wasn’t the person who lived the workflow.
Now flip the model. When the business expert is the person directing the software, the translation loss disappears. The restaurant owner who knows her food costs to the penny. The contractor who knows exactly how his change-order process breaks down. The claims manager who knows which three data points actually predict a problem file. These people have spent careers accumulating knowledge that no software vendor will ever fully capture.
They were never the problem. The $150-an-hour translation layer was the problem. And it’s gone.
What this looks like in practice
I’m not speculating. I run a P&C insurance technology company, and I build production software with Claude Code daily — data pipelines, document processing systems, integrations between platforms that were never designed to talk to each other. Work that would have taken my team weeks now takes days. Work I would never have greenlit at all — internal tools too small to justify developer time — now just gets built.
Earlier this year I worked with a small business owner in the healthcare cost-containment space. Within a very short time of launching his first AI-assisted system, he’d landed a new customer. Not because the technology was magic — because his expertise, which had been locked in his head and his spreadsheets, was finally embedded in software that worked the way he worked.
That’s the pattern I believe we’ll see repeated thousands of times over the next few years: business owners who are experts at their business, designing and creating their own software.
The honest part: what you still can’t skip
If this article ended here, it would be a hype piece, and I’ve been in this industry too long to write one of those.
AI coding tools are extraordinary, but they are power tools, not autopilot. Here’s what a business owner building their own software still genuinely needs help with:
Architecture decisions before you build. The AI will happily build whatever you ask for — including a design that paints you into a corner six months from now. Knowing how to structure data, when to split systems apart, and what will and won’t scale is judgment that comes from experience, not from a prompt.
Security and data handling review. Your customer data, payment information, and compliance obligations don’t care how the software was written. AI-generated code can look perfect and still leave a door open. Someone with security experience needs to review anything that touches sensitive data before it goes live.
Deployment and infrastructure. Getting software from “works on my laptop” to running reliably where your team can use it — hosting, backups, updates, monitoring — is a separate skill from building it.
A safety net when things break. Software breaks. When it does, you want a relationship with someone who has seen that failure mode a hundred times, not a support ticket queue.
Notice what’s not on that list: writing the software. That part is now yours.
The new division of labor
The old model: you pay a development shop $200,000 and they own your software knowledge.
The new model: you own your software — you designed it, you understand it, you can change it — and you pay an experienced engineer a fraction of the old cost for the guardrails: architecture guidance up front, security review before launch, deployment help, and a review relationship as you grow.
This is the biggest transfer of power from software vendors to business owners I’ve seen in my career. The businesses that seize it will have software shaped precisely to how they operate — a genuine competitive advantage their SaaS-bound competitors can’t buy at any price, because it’s built from expertise only the owner has.
We are at the very beginning of this shift. The tools arrived before the awareness did. If you’re a business owner with deep expertise and a list of things your current software can’t do, the constraint you’ve lived with your whole career just lifted.
The question isn’t whether you can afford custom software anymore.
It’s whether you can afford to keep renting someone else’s.
Datos helps business owners design, build, and take ownership of their own custom software — architecture up front, security review before launch, and a relationship for when things break. Get in touch or read more about AI consulting.