Does AI make custom software cheaper? What the numbers say

Yes and no. AI genuinely makes software development faster, and on well defined new projects you see that in the price too. The spectacular discounts some agencies promise, up to 60 or 70%, appear in no serious research anywhere. Below we put the claims next to the studies, so you know what to expect in 2026 when you have software built.

What do agencies promise about AI and price?

Some agencies promise that AI makes your software up to 60 or 70% cheaper. Those are marketing claims with no published method behind them. Providers in the Netherlands in particular dare to go far, with percentages you will not find in any study. Invenix writes that developers using tools like Cursor and Claude Code "deliver 30 to 50% faster than two years ago" and that a €60,000 project from 2023 can now be done for €25,000 to €40,000. Bespoke Automation goes as far as claiming software "built 70% faster" and "priced 60% lower". The research literature paints a calmer picture.

What does the research say about AI and coding speed?

The research shows that AI mainly speeds up small, new and well described tasks, and on complex existing code it sometimes slows things down. The gain is real, and it is smaller and more erratic than the marketing suggests. The four soundest sources, with what they actually measured:

• One small, sharply defined task: better than half the time. In a controlled GitHub experiment, 95 developers built a small web server; the group with an AI assistant finished in 71 minutes, the control group in 161. That is 55.8% faster, on a single new and well described assignment.
• In production at large companies: 26% more tasks completed. Field experiments at Microsoft, Accenture and a Fortune 100 company covering 4,867 developers found 26% more completed tasks. Junior developers gained the most; for seniors the gain was limited.
• The disappointment: experts on their own codebase worked slower. In the METR study, highly experienced open source developers took 19% longer on tasks in their own large projects when they used AI. The striking part: they believed they had been 20% faster. Perception and measurement drift a long way apart.
• Across teams: AI amplifies what is already there. The DORA report 2025 from Google (around 5,000 professionals) sums it up like this: strong teams get better at it, struggling teams see their problems magnified. Release stability remains a point of attention where AI adoption is high.

Add to that the Stack Overflow survey 2025: 84% of developers use AI or plan to, while 46% distrust the output and 66% name "code that is almost right" as their biggest frustration.

So where does building software actually get cheaper?

On new and predictable work: an MVP, prototypes and standard integrations. That is where AI speeds up the routine work and the price drops noticeably. For your project that means:

• An MVP or first version gets faster and therefore cheaper: a lot of standard work (screens, forms, integrations done by the book) lends itself perfectly to AI assistance.
• Prototypes and experiments now cost a fraction of what they used to. Testing an idea before you invest is more achievable than ever.
• Standard integrations (payments, accounting connections, Peppol) go more smoothly because AI speeds up the documentation and the routine work.

Where the gain is small or even negative: architectural choices, complex existing systems, edge cases and maintenance. Those are about understanding and deciding, and that stays human work with review. We bundle that mix of AI speed and human control under AI orchestration.

What does this mean for the price of your project?

Our price ranges (€8,000 to €25,000 for an MVP, €25,000 to €75,000 for a full application) are already prices from the AI era: they assume the routine work runs faster. You mostly see the gain in what you get for that budget: more finished functionality, faster first versions and cheaper experiments. What does not drop: the time for thinking, testing and connecting to the messy reality of your business.

If an agency promises you a 60% discount thanks to AI, ask what happens to the time it saves. If the answer is "less review and fewer tests", you pay that discount back later in bug fixes.

How do you recognise an agency that uses AI sensibly?

Ask three questions before you sign: who checks the AI code, what the speed claim rests on, and what happens after launch. The answers quickly give away whether an agency uses AI as an accelerator or as a sales trick.

1. "Who checks the code that AI writes?" The right answer contains words like review, tests and a human being who stays responsible.
2. "What do you base your speed claims on?" Measured numbers from their own projects are a good sign; the METR study shows that gut feeling systematically overstates things here.
3. "What happens after launch?" AI-generated code that nobody understands becomes dead weight the moment something has to change.

---

Want to know what your idea costs in 2026, with AI acceleration where it makes sense and human work where it has to be? Put it to us and we will split the estimate honestly: what AI speeds up, what it does not, and what that saves in your budget. You get a price we can justify, and no promised miracle discount.