What causes an AI agent to become less reliable?

Reliability degrades whenever something the agent depends on changes, and agents depend on more moving parts than most software. Teams edit prompts, swap or update tools, and change the data the agent reads, each on its own schedule and each capable of shifting behavior. The model itself can change too: providers update models behind the same API, sometimes without announcing it, so an agent can get worse without anyone on the team deploying anything. Complexity compounds the problem. As agents take on more tools and longer chains of steps, there are more places for a change to land and more steps where one bad decision feeds the next. None of this shows up as an error. The agent still returns a completed response, so a decline in quality has to be measured against what the agent used to do, not read off an error rate.