What AI really changes in a software project

Every week a business owner asks us the same question: if AI writes code now, why does software still cost money? It is a fair question. The honest answer is that AI changes where the time goes in a project, not whether you need people who know what they are doing.

In our projects, AI tooling makes a clear difference in a few specific places:

• Reading and documenting existing code. A legacy codebase of 100,000 lines used to take weeks to understand. Now we can map modules, data flows and hidden business rules in days.
• Writing the boring parts. Forms, admin screens, API clients and test scaffolding take a fraction of the time they used to.
• Tests. Generating a first set of tests around existing behaviour is fast, which makes migrations much safer.
• Prototypes. A clickable prototype with real data can be ready in the blueprint week instead of after a month.

Some parts of a project are just as hard as they were five years ago:

• Deciding what to build. AI cannot sit in your warehouse and notice that planners keep a paper list next to the screen.
• Data quality. If your customer records are messy, a faster developer only gets to the mess sooner.
• Integrations with old systems. The slow part is usually getting access, credentials and a test environment, not writing the code.
• Review and responsibility. Every line that goes to production still needs a senior engineer who understands it.

For a typical SME web app, we see the build phase shrink by 30 to 50 percent compared to a few years ago. Discovery, design and testing with real users stay roughly the same. So the total project gets cheaper and faster, but not by the 90 percent some headlines suggest.

The bigger change is what becomes affordable. Projects that used to be too expensive for a company of 30 people, like replacing an ageing ERP module or building a proper customer portal, now fit in a normal budget.

If you are talking to a development partner, ask them:

• Which parts of the project do you use AI for, and which not?
• Who reviews AI-written code before it goes live?
• Do the savings show up in the quote, or only in your margin?
• Who owns the code and the prompts when the project ends?

A good partner answers these without hesitation. If the answer is vague, the savings will probably be vague too.

We call ourselves AI-native because we build every project around these tools, not because the tools do the work. The result is smaller teams, shorter timelines and more budget left for the things that actually decide whether software gets used: understanding the problem and getting it in front of real users early.