From idea to finished tool in two hours: what we learned while developing with AI#

Those who programme with AI support can build a functioning tool in a short amount of time. The key? Don’t start coding right away – first describe exactly what you want.

What was the goal?#

We wanted to test whether text style adjustment could be made easier.

An earlier project had copied writing styles from sample texts (such as those by Goethe or from business letters). It worked, but it was complicated. The new idea: users set the desired characteristics themselves – using sliders.

What can the tool do?#

It transforms texts into different writing styles:

Adjust tone (formal or casual, emotional or factual)
Control complexity (simple language or technical language)
Add creative elements (irony, poetry, humour)
Define target audience (children to experts)

A total of 34 sliders in 7 categories. Plus 23 ready-made templates for typical cases such as ‘job applications’ or ‘social media’.

How did we develop it?#

In two phases:

Phase 1 – Preparation (~60 minutes): We discussed with the AI what the tool should be able to do. Goals, functions, appearance, technology – everything was defined.
Phase 2 – Implementation (~60 minutes): The AI then wrote the code in one go.
Total effort: 2 hours for 2,000 lines of code
Result: 12 files, executable with one click

Why did it work so well?#

Because we spent more time on preparation than on implementation.

The description ended up being 1,400 lines long – almost as much as the code itself. Sounds excessive? It was worth it. We hardly had to make any corrections afterwards.

Initial tests show that the tool also works for fine adjustments, not just for drastic style changes.

Key insights#

1. Preparation beats correction

If you tell AI to ‘build me a tool for X’, you’ll get something – but probably not the right thing. If you describe exactly what you want beforehand, you’ll save yourself a lot of correction rounds.

2. AI helps with planning

We didn’t write the description on our own. The AI asked questions, suggested alternatives and pointed out gaps. The result was better than if we had done it ourselves.

3. Less is sometimes more

The previous tool had a similar amount of code, but could do more (analysis AND conversion). The new tool can only convert – but it is easier to understand and use.

What can we take away from this?#

A good description is more important than getting started quickly.
AI isn’t just for coding – it also helps with thinking.
Simpler concepts often lead to better results.

Conclusion#

✔ 2 hours from idea to working tool

✔ 1,400 lines of preparation for 2,000 lines of code – the ratio shows where the real work lies

✔ Developing with AI means less typing, more thinking

This is part of a series on software development with AI support. The focus is on what can be learned from such projects – not just on the results.