How AI reduces development costs by 2900%: one week versus one year of development

It has been about thirty years since I wrote my first script. From experience, I can tell you that being able to program is extremely useful when starting a business, especially if you want to build a cloud communication platform.

For example when:

• A CRM package had to be linked to our telephony platform
• The first generation of our invoicing system was written
• CRM data had to be used to write effective quotes

Basically, I could do it all myself. But a good programmer? I never became one.

Why I stopped programming

I write code that works, but nothing more than that. Good software is much more than just working code. A solid code architecture, comprehensive documentation, and good programming principles are essential. Only then can you build systems that are easy to maintain and expand.

That is precisely why I haven’t written code for over ten years. Or rather, I am no longer allowed to write code by the many more talented developers around me.

The AI experiment that changed everything

But as you may know, I’m quite interested in AI. I like to explore the limits of what’s possible. That’s why I decided a while ago to investigate how mature AI has become in the field of programming.

The result? I was shocked.

After seven days, I had a highly advanced, production-ready system that allowed me to route incoming customer calls based on factors such as current CRM data. But I was really shocked when I had someone estimate how much time a mid-level developer would need to do this.

The answer: 380-560 hours. At that point, I had only spent 40 hours on the project.

The actual development costs: an eye-opener

The footnote to that estimate was even more telling. The 380-560 hours was possible because this project was “exceptionally well documented and the programmer would know exactly what to build.”

But that documentation—and the thinking behind the project—came about organically during programming.

So the follow-up question was: how much time would this take in a regular development process? If you go through the complete cycle of:

• Business and design thinking
• Usability and user interface design
• User validation
• The actual programming

The answer: 1155-1750 hours. Yes, you read that correctly, that’s 10-12 months. Almost a year.

With the help of AI, I had built something in one week that we could immediately validate in the market.

The “too expensive” project that suddenly became feasible

The challenge: smart call routing

At Voys, we connect customers with the right person in the organization as effectively as possible. To ensure this, the following information is required:

• What is the customer relationship?
• Who has had historical contact with whom?
• Is the relevant colleague actually available?

That is why we developed Voys Reach and Voys Pulse with CRM integration.

The fundamental question

I wondered: what is the best predictor for the answer to the question of who the customer really wants to speak to?

• The phone number the person is calling (e.g., Finance department)?
• Recent conversations that have taken place?
• The open support case in the CRM system?

Why this project was put on hold

Of course, I could have asked my colleagues in development to create a system for this. But that would have taken an enormous amount of development time. And the value for our average customer would have been limited in relation to other priorities.

So I walked around with this question for years, without getting an answer. Until AI changed that.

From prototype to production

The data from this research project is fundamental to efficient call and chat routing. That’s why I wanted to test how far I could get with AI.

The answer is simple: you can build a highly usable, production-ready system in a fraction of the time.

At the moment, I’m still calling it a prototype. But how big is the step from prototype to actual product?

What this means for the future

It has been a long time since I learned so much in 14 days. And the consequences of this development are difficult to foresee.

Blog series ‘Programming with AI’

This is the first article in a four-part series on programming with AI. You can read the other parts here:

Part 2: Programming with AI: clever cheating

Part 3: 19 tips for successful programming with AI and Claude Code