My AI journey as a developer

The starting point

At the company I work for, most of our systems are built on Oracle APEX, with PL/SQL as the backbone. We make use of JavaScript and CSS of course for Oracle APEX and the custom components we build for it. I figured it would work for the JavaScript and CSS part but was kind of expecting it not to be very good in PL/SQL.

Go wild with AI

My company basically gave all the developers Claude accounts, some tutorials, and online courses and told us to “go wild.” I was still quite sceptical, but at the same time, it was really interesting and fun to play with.

To give a bit more context about the company, the company I work for provides a SaaS solution used by thousands of people every day. For this reason, there is a big focus on stable, clean and maintainable code. From all the stories I heard about AI I figured that this was going to be one of the bigger issues.

The eye-opener

As a first step, I was very interested in figuring out if it could understand the system, so I decided to upload some architectural documentation and some functional descriptions of how the system works. It managed to give me a very good description of the system itself and its design. I wasn’t particularly impressed, but it was a nice starting point.

From scepticism to blown away

Then I decided to take it a step further and start uploading quite a bit of code and decided to start building a story. That was the moment all of my scepticism completely disappeared. It was fairly clean code, it knew what it was talking about, and it actually worked. It gave me exactly the right things, it followed coding guidelines which it got from both the architectural documentation and inferred from the already existing code. This completely blew me away and made me realize that this AI thing might be here to stay.

Practical use case

Currently I use it in quite a bit of the code I am writing, and I notice that it has been helping me to write better code. As an example, I recently used it to help write an integration system with a new partner of ours.

What I did was write down what the flow of the system should be, which API calls should be used back and forth, and what code I expect, in very simple pseudo code, and then I give it to AI to start filling in all the details. Within 20 minutes, it has all the code, and this code follows all the guidelines of the company. From this point I just do a sanity check to see if it has missed anything.
Once I was done, I simply used it to generate some documentation, a few UML diagrams (this does require some very precise instructions, or it will generate way too much). This saves me quite a bit of time.

The lesson

A few things I noticed while using AI:

• Giving it enough context is very important, both technical and functional.
• Giving it specific instructions helps a lot while writing code, don’t be too detailed though, keep it very high level.
• Don’t be afraid to start over in a new chat if it really does things you don’t want it to do.

Conclusion

Looking back, I understand why I was sceptical in the first place. Most of what I had seen about AI before this didn’t convince me it belonged anywhere near production code. But once I actually gave it enough context and the right kind of instructions, it turned into something I can somewhat rely on now.

AI helps me to keep my code maintainable and documented, it can even do reviews to check if I have missed anything / made any logic mistakes which can be very helpful.

I don’t think AI replaces the developer, it is not even close after my experiences with it. I still check everything it gives me, and I still need to know what good code looks like to catch it when something is off. But it has made me faster, and in some cases, it has made my code better than it would have been otherwise.