For most of the last thirty years, custom software had brutal economics for small organizations. Even a modest internal tool meant weeks of billable development — so anything that couldn't justify an agency budget stayed a spreadsheet, a binder, or a someday. Whole categories of usefulness were priced out of existence for everyone below the enterprise line.
That line just moved. Significantly. And because "AI" on an agency website usually signals hype, we want to explain the mechanics plainly enough that you can judge them yourself.
What actually got cheaper
Software development has always been part invention and part production — and the production share is bigger than outsiders assume. The boilerplate, the plumbing, the thousandth login form, the data table with sorting and filtering, the API integration that's 90% like the last twenty API integrations anyone wrote. This work isn't hard; it's voluminous, and volume was what you were billed for.
AI-assisted development compresses exactly that layer. An experienced developer working with modern tools produces the routine majority of a system dramatically faster than in 2019. Not because the machine is clever — because the work was always mechanical, and mechanical work is what machines are for.
The result: a focused tool that would have been a five-figure engagement a few years ago can now often be delivered at a price a small business can justify against the hours it saves. Not free. Not magic. Just — finally — proportionate.
What didn't get cheaper
Every part of the job that was never about typing is untouched, and this is the part the hype omits:
- Scoping. Deciding what the thing should be — and how much of it not to build. A generator will happily produce the wrong product ten times faster. Smallest useful version, always.
- Architecture. The decisions that determine whether the system survives contact with year two: data models, boundaries, what happens when a feed rots or a dependency dies.
- Security and correctness. Generated code is confident code. Confidence is not correctness, and nobody's model is accountable for your customer data.
- Knowing when not to build. The most valuable sentence in this business is still "an off-the-shelf product already does this — buy it." No tool volunteers it.
In other words: production got cheap, judgment didn't. Which means the judgment is now most of what you're paying for — and you should evaluate providers accordingly.
What this means if you shelved something
Dig it back out. The quote you got in 2019 is not the price today — but more importantly, the shape of what's worth building has changed. Projects that were sensibly rejected at old prices — the internal calculator, the customer portal, the automation that saves ninety minutes a day, the small product for a small niche — are frequently sensible now.
Two honest cautions. First, cheaper to build also means cheaper to build badly — the market is filling with fast, fragile software, and the difference won't show until it breaks. Second, some shelved ideas deserve to stay shelved, and lower cost isn't a reason to build a thing nobody needed. The evaluation — is this worth building at today's prices, at what size, or at all — is precisely the conversation to have before money moves. Ours comes with a written, plain-English estimate, and "don't build this" is a real possible outcome. That advice is free.