AGI and SRE

Thursday, 15 October, 2026 - 15:50–16:35

Alex Palcuie, Anthropic PBC

For two years the SRE industry has discussed how AI models are changing our jobs. What nobody has seriously discussed is what happens when a model becomes at least as good as an SRE at the core job like holding the pager, mitigating incidents, and even reasoning about production. This talk argues that scenario is plausible within your career. Frontier labs are already publishing reports on how their models help build their successors, a process the industry calls recursive self-improvement (RSI). The rest is about how to manage the transition and what SRE becomes once the models are good enough.

I'm a Member of Technical Staff at Anthropic, where I keep Claude inference reliable. That means the unenviable job of fixing Claude without Claude when it goes down. The Register has covered me twice, once for a QCon talk on using AI models for incident response, and once for a global GCP outage I inadvertently caused. Previously I spent eight years as an SRE on Google Compute Engine (GCE), where I was a founding engineer of Cloud IRT, the team for massive GCP outages, and served on Tech IRT, which runs Google-wide incidents.

Presentation Video