Analyst Chat #316: Integration Debt & the Vibe Coding Trap in Identity Management

The Lotus Notes analogon was the one that I was waiting for.

The other is I like that sentence, integrations or orchestration that work until they don't.

So this is the part that everybody, I think, who is active in IBM has seen before.

Of course, not in their current project, but in the one before.

So it's really highly relying also on people, on heads, on people who did it, who understand what's happening.

And if they leave, that's clearly also a sign of not having the proper governance of or over the orchestration itself.

And I think that's something where we now have technologies and ways and means to just get better on that.

You've mentioned already the ways how we do coding.

But also quality assurance might be an aspect where new technologies, of course, I'm thinking of artificial intelligence or LLMs or supporting tools that can just support us in getting better unless or as long as we do not trust them too much, right?

Yeah, but I think that there's this promise so we can use AI to code faster, which usually means we don't understand what exactly is happening so we just trust it we can then use the same or other llms to do the quality assurance of the code we can even use that to document code so i'm i'm from from a generation where when i started coding it was a not only a good practice but also a common practice to comment on the code so there were not only the lines of code there were also all the comments Around that, that was something I think which disappeared a bit over the years.

AI obviously can document.

The problem is AI can check for bugs, but it doesn't understand what is really meant, what is the purpose of the code.

So you can give it a task and say, okay, I need code for that.

But the other way around, saying this is the code, what is this code doing, is way more complex.

And obviously, AI is not really good in it.

When we look at the news of just the last weeks, with all this, we tested this LLM and it escaped from that environment and it did rogue things.

Hey, if AI would be good in understanding what could be the result of running certain code, what the code could do, then this would not happen.

So AI, the large players prove that they can't do that yet.

So it's good in documentation.

It helps us to a certain extent.

But if you end up with hundreds of orchestrations that are around and try to figure out why were they built, what's the purpose of that documentation, AI won't solve this problem for you.

So we still need a governance, a framework where we do all the orchestration, where we implement the good practices in the organization.

All this needs to be done.

And by the way, people will like that if we do it right.

If it's not just about being restrictive, but helping people doing their job easier.