Companies ship more code today than ever, thanks to agentic coding tools. For product teams, this accelerated pace of development is a boon. But for site reliability engineers (SREs), the influx of new, machine-written code is not.Why? Because when something breaks, no human fully understands how the underlying system works. Sorting out what went wrong, how to fix it, and how to avoid the problem in the future can be both difficult and time-consuming. All while time constraints remain static, as downtime is no more welcome today than in the past.You can easily burn out your humans by tasking them with sorting out how to fix what agents built; that’s backward.If agents are writing code and humans are having a hard time keeping up, is the solution to fight fire with fire — to deploy AI agents to sniff out root problems in misbehaving systems? Yes. The good news is that as AI models have improved, the capability to execute root-cause analysis is improving, though the OpenRCA benchmark remains far from saturated. Thanks to those performance gains, Sam Farid and Nate Heinrich of Chronosphere, a Palo Alto Networks company, say that agents are the way forward. And the pair think that your company should try to build an AI SRE in-house — before exploring vendor offerings.Chronosphere offers an AI SRE product, so why are its forward-deployed podcasters arguing for homegrown alternatives? Heinrich says in this latest episode of The New Stack podcast that he tries to convince people to build their own agents because doing so is a useful way to collect and organize information about how a company’s systems work. The result of the exercise is a Markdown file that agents can use as critical context when sniffing out root causes; the process of collecting the information is both the journey and the destination for companies willing to put in the work.Naturally, Chronosphere believes that after a company tries to build its own in-house SRE agent, it’ll discover that it needs a telemetry service to collect logs and traces, and an observability tool to store and correlate that information. Which it offers. Along with a ready-built AI SRE. But I really appreciated sitting down with two technologists who weren’t trying to tell me that their product, and only their product, is the way forward. Hit play, and enjoy. You just might find the courage to start mapping your internal systems for eventual agentic consumption. After all, don’t you want help when something breaks in the early morning hours, and you didn’t write the code that just fell over?The post Why your company should (try to) build its own AI SRE appeared first on The New Stack.