Owners ask a sharp question now: if AI writes code, why does custom software still cost what it costs? It deserves a straight answer, because both halves of the easy version are wrong. AI has not left development untouched, and it has not made a custom system a fraction of the price. This is what has actually changed in how we work, what has not, and how it shows up in what you pay.
What got faster
AI is in our workflow every day, as a tool inside a professional process. The gains are real and they are specific:
- Research and options. Evaluating an integration, a library, or an approach used to take a day of reading. It takes an hour of directed questions now.
- First drafts of code. Boilerplate, data models, the routine parts of an interface, test scaffolding. The parts of the job that were typing more than thinking.
- Tests. Writing them was always the step that got squeezed. AI generates thorough ones cheaply, so coverage went up rather than cost coming down.
- Documentation. It is written as we go now, because it costs almost nothing to produce from the work itself.
- Reviewing AI’s own output. We use AI to review what AI produced, a second pass that catches a class of mistakes before a person spends time on it. A senior engineer still reviews all finished work before it ships. The tools direct nothing; people do.
Net effect: a project that would have taken a team six months a few years ago takes them noticeably less, and the quality is higher for the same effort, because the tedious parts are no longer competing with the careful parts for attention.
What did not get faster
The hard parts of custom software were never typing.
- Understanding the business. What the system has to do, for whom, and what it has to connect to. This is discovered in rooms with your people, and no model shortens it.
- The integrations. Your ERP’s API is as undocumented as it was before. The data in your CRM is as messy. Making two systems agree on what a customer is remains a human problem.
- The decisions. What to build first, what to leave out, when the scope should flex. Judgment does not compress.
- Testing against reality. An automation or a screen has to be tried on real work by real people, and their objections have to be heard and acted on.
- Change in your organization. The rollout, the training, the week when the old way is switched off. Unchanged.
Roughly speaking, the work that got faster was the work that was already the cheapest part of a project. The work that stayed slow is the work that was always the expensive part.
The trap on the other side
There is a new way for a project to go wrong. Software generated quickly without engineering around it: no architecture, no tests, no security review, no one who understands it. It demos beautifully. It then falls over on the first real data or the first real user, and it cannot be fixed because nobody knows how it works. We see the results, because those systems come to us for a System Evaluation and a takeover. Speed without discipline is not cheaper. It is deferred cost with interest.
What it means for your budget
Three things you should expect from a firm in this environment, and what you will see from us:
Lower cost for the routine parts, honestly passed on. Our published ranges reflect how we build today. The floors on the pricing table are where they are because the first release of a well-scoped system is genuinely cheaper to produce than it was.
No discount on the planning. The Validation, Design & Planning step is worth more now, not less, because a team that can build fast can build the wrong thing fast. The mockups, the requirements, and the roadmap are what point the speed in the right direction.
The same disciplines, priced in. Architecture, testing, security, documentation, and a person accountable for all of it. These are what make a system safe to depend on in year three, and they are the part of the budget you should be most suspicious of a competitor cutting.
If a proposal is a small fraction of ours for the same scope, the difference is almost never AI. It is one of the disciplines above, missing. Ask which one.