The Future of Observability with OpenTelemetry

Chapter 3. The Limitations of Automated Analysis

Automation is beginning to sound amazing, but here’s an important reality check: at the end of the day, computer analysis will not be able to tell you what is wrong with your system or fix it for you. It can only save you time.

With that in mind, I want to stop with the praise for a moment and acknowledge some limitations of automation. I do this because there will be quite a lot of hype around combining artificial intelligence and observability. This kind of hype leads to amazing claims along the lines of “AI anomaly detection and root-cause analysis will fully diagnose your problems!” and “AI operations will fully manage your system!”

Beware of Hype

To derail this hype train, I want to be clear that these kinds of dreamy marketing claims are not what I am claiming modern observability will provide. In fact, I predict that many claims related to intelligent AI problem-solving will mostly be a lot of snake oil.

Why won’t AI solve our problems? In general, machines cannot identify “problems” in software because defining what counts as a “problem” requires a form of subjective analysis. And the kind of real-world AI we are talking about cannot perform subjective decision making with any degree of accuracy. Machines will always lack sufficient context.

For example, let’s say that a release slows down performance. But it also contains a really important feature that’s number one on every user’s wish list! In that case, the performance ...