Include maintenance cost in your business case if you build your AI agents yourself

I know that many people believe that it is easy to set up AI agents. And they are often right. Today, with the help of Claude or ChatGPT or some of the other available AIs, it is probably possible in one prompt to get something up and running.

The problem is that at some point it will crash. When it crashes, someone needs to fix it.

So maintenance for IT tools like an AI agent is the same as for any machine. There is reactive fix and there is preventive maintenance.

The reactive fix is because something stopped working and it requires immediate attention. This is obvious and the important thing here is that for the AI agents, roles and responsibilities are clear of who needs to fix it when one doesn't work.

The preventive maintenance is a little more complicated. Maybe the surrounding processes are changing or some of the programming frameworks used for the build of the AI agent is getting outdated. In this case, the AI agent will still work but, a technical debt is slowly building up and the agent will become more of a burden.

For example in one of my previous jobs, my team built plenty of automation scripts with lot of positive impact. But as time passed, the scripts were built with different newer structures which led to a technical dept and general complications when bugs appeared.

The trick here is not to avoid building the AI agents but rather to include the cost of maintenance in your business case.