A year ago, writing code took up most of the budget for a typical website or internal tool. That is no longer true. AI tools now write a large share of the code in our projects, and we see it in the market too: quotes for projects like a new marketing site have dropped a long way in a year.
So does AI make development cheaper? Writing code got much cheaper. Deciding what's worth building didn't. Both halves matter if you are planning what to spend your budget on.
Writing the code was never the hard part
For most business software, writing code was the routine part of the job. Forms, listings, integrations and admin screens follow patterns an experienced developer has written many times. That routine work is what AI does well, and it now produces it in minutes rather than days.
Our CTO, Alexander Sundli-Härdig, describes it as an 80/20 split: getting a solution 80% of the way is easy now, and the last 20% still takes people who know what they are doing. That last part is the edge cases, the security, the performance under real traffic, and whether the thing actually does what your business needs.
What it makes possible: tools that used to fail the business case
When the build costs less, the list of things worth building gets longer. Internal tools, integrations and small applications that used to cost more than they would save can now pay for themselves. We see clients spend their budgets on exactly that.
- Steni wanted to test an idea on its own product data. Nine days after development started, they had a working prototype to try on real data.
- Moxtex runs a business that off-the-shelf order management didn't fit, so we built a system around how they work. The first version was built within a week.
- Veterinaria translates its entire site from German into English to reach the many people in Austria who don't speak German.
- At Represent, we built our own reporting tool to automate our weekly client reporting. Before, we could not justify the cost of building it.
Speed matters less in these cases than the business case: each of these ideas now pays for itself. If you have a list of ideas you shelved because the quote was too high, it is worth revisiting.
What didn't get cheaper: deciding, checking and owning the result
AI will build what you ask for, quickly, including the wrong thing. With the build no longer the main cost, the expensive mistake is spending a budget on features nobody needed. Helping clients choose what to build, and what to leave out, is now a bigger part of our work than it used to be.
Checking takes longer than writing. In our projects, testing and verifying now take more of the hours than producing the code does. Does it behave with real data? Does it match the design down to the details? Does it hold up on a slow phone? AI still gets details wrong, and it rarely tells you when it has.
Earlier, budgets were stopping ideas from being built. Today, building is cheaper. Now it is even more important to pick the right ideas that deliver a good ROI.
Then there is security. Code that works can still expose data or fall over under load, and an AI tool won't flag that on its own. We review every change before it goes live, with automated security checks on top.
How it has changed the way we work at Represent
Our developers write less code by hand than they did a year ago. More of their day goes into describing tasks precisely, reviewing what AI agents produce and deciding how things should be built. As one of them put it, every developer now has to think more like an architect, with AI doing the work a junior developer used to do.
A typical setup on our projects: a developer describes the task, an AI agent implements it, a second agent reviews the work, and a developer reviews it again before anything is merged. We cap how many times the agents go back and forth because after a few rounds, the result tends to worsen. Merging stays a human step.
We also keep each project's history in its documentation: decisions, rules and things that went wrong before. The effect is noticeable: it is the same AI, but the results improve significantly when it knows the project's history.
Design is changing in a similar way. On suitable projects, our designer sets up a design system with AI tools and then works more like an art director, which takes a fraction of the hours a traditional design phase did.
Some things we keep strict. Client work only runs on business AI accounts with data-handling terms in writing. AI agents can't merge code, delete data or touch production systems on their own. And some clients don't allow AI in their projects at all, for regulatory or group policy reasons. For them we work the traditional way, to the same standard.
Not every process has changed yet, either. Much of what we do with AI is still the old process done faster. The bigger shifts come when work runs on a schedule without anyone starting it, such as routines that gather what needs attention each morning. We are building more of those.
Where AI doesn't help yet
AI-assisted development isn't the right fit for everything, and we would rather say so up front.
- If your organization has strict rules on data and AI tools, the traditional approach is the right one, and it still works well.
- A generated prototype can look finished long before it is. It is a good way to test an idea, and a weak basis for decisions about the final product.
- Estimates are still settling. The build is faster, but how much faster varies by project, and we say so when we quote.
What this means for your budget
We see AI as a way to get more out of a budget: more of the ideas on your list, more testing and more improvements after launch. We also advise leaving part of the budget unplanned, because once a project starts you will think of things you want that weren't in the brief.
If there is an idea you shelved because the quote was too high, it may be worth another look. Read more about how we work with AI, or get in touch and we will talk it through.