IMHO it's perfectly fine to use AI to find the exploits. I was just emphasising that the LinkedIn article reads like a direct and unsanitized wall-of-text-ish output of a chatbot. If you're happy with that, that's also ok.

Be honest, did you use AI to write that story down?

I love that the mascot of Xogot is a cute axolotl lol.

I hope that, if and when the Java templates arrive, we can have the best of both worlds: safety and readability.

I think that's a good analogy, but in case of the agentic setups, paradoxically we may have stable APIs (tools) but have to deal with a non-deterministic brain (the LLM). The risk of incosistencies is high and the propension for security in this new MCP setups is low...

Only then, to me, adding tests is then useful. Only when the damn shape of the things come together. Adding tests is even easy then.
In my career, I never saw that tests written before can actually help, except *extremely* trivial examples. Whoever wants to sell that idea to you, is blatantly lying.

I just write tests alongside code only when I have to implement complex algorithm and encapsulate it into a class or similar. That's a case when I think "I need some tests before moving forward".
I feel satisfied with my code when it's already reworked to the max to cleanliness and layering.