Why “paste the log into the agent” fails for root cause analysis

A failed CI pipeline leaves a large log and a simple question: why did this fail? The tempting approach is to give the entire transcript to an agent with a large context window and ask it to find the