Weekly notes: Agents only look to logs 30% of the time. Very interesting data about how an agent given a general question and multiple data sources chooses where to look ...

• What about juniors?: When writing code has been commoditized and made easier what then do people getting started focus on? They’ll probably have to think more about the why of building something, the users, the how tos of making stuff happen, etc.
• Context, Know-How and AI SRE: What Is the Other 70%?: Finding how to get the most out of agents from the production support frame.
  • Root cause analysis (first thing we tried) vs general question + answer bot across production (hard but there’s value here)
  • Diverse sources of context
  • How would you answer a question like “What happened yesterday?” after an incident when a system is comprised of hundreds of servers and processes. Where does an agent begin?
  • Another insight is the documentation we write might not actually be as helpful as we might think (eg out of dozens of pages in a wiki / runbook, the useful information to an agent was 3 lines)