Speaker
Abstract
Can you throw an LLM at a production incident and expect useful results? A candid look from someone who runs a distributed AI system and reaches for Claude before reaching for a dashboard. Surprises, failures, and why the answer matters for every engineer carrying a pager.
Interview
It's a field report on using LLMs for incident response. I run a production AI system and these days I reach for Claude before I reach for a dashboard. It's still taboo to say this, and sometimes it would have been better to just open the dashboard, but I want to discuss when it is and when it isn't.
If you decide to skip this because it's yet another talk about AI, I totally understand. I sometimes think there's too much talk about AI and not enough doing. This one's a field report, I'm definitely not going to try and convince you that Claude will solve all your problems.
Loss of control feels very scary. What once felt like a comfortable on-call rotation where you knew all the nooks and crannies now includes a large language model that sometimes finds the issue faster than you can and sometimes feels like an overconfident junior.
Curiosity for experimenting. We're all learning together.
The audience has been paged at 3am and has worked with mission-critical systems. I can skip the intro and go straight to the interesting part.
QCon London 2026 is a three day conference for senior software engineers, architects and team leads. An international program committee of working engineers selects every session. Patterns and practices, not products and pitches.
From the same track
Tuesday 17 March
10:35 Windsor (5th Fl.) Session Sociotechnical Leadership Orienting, Understanding, Playing, Thriving: Debugging your Organisation Hazel Weakly Fellow @Nivenly Foundation; Director, Haskell Foundation; Experienced Leader Focusing on Organizational Change, Developer Experience, and Resilience Engineering Debugging is both an art and a science. But more than that, it's an activity undertaken with deep intention: to understand and improve your systems. In the purely technical realm, we have an extraordinary range of tooling and techniques that can help us tackle this problem. 11:45 Whittle (3rd Fl.) Session Distributed Tracing How Eve Online Leverages Head Based Sampling to Observe "Fun" Nicholas Herring Technical Director, Eve Online @CCP Games, Refiner of Internet Spaceships and Explorer of Feral Gordian Knots of Python A unique pattern in video game software is real-time interactions to express the personality of users. Here we will talk about how we instrument the universe of New Eden to identify the traffic that matters, even the "fun" parts! 13:35 Rutherford (4th Fl.) Unconference Unconference: Debugging Distributed Systems 14:45 Fleming (3rd Fl.) Session Can Claude Fix Itself? Using LLMs for Incident Response Alex Palcuie Member of Technical Staff in AI Reliability Engineering @Anthropic, Previously Staff Site Reliability Engineer on Google Cloud Platform Can you throw an LLM at a production incident and expect useful results? A candid look from someone who runs a distributed AI system and reaches for Claude before reaching for a dashboard. Surprises, failures, and why the answer matters for every engineer carrying a pager. 15:55 Mountbatten (6th Fl.) Session Observability Wrangling Telemetry at Scale: A Guide to Self-Hosted Observability Colin Douch Site Reliability Engineer @DuckDuckGo Observability is supposed to help you tame complexity, but your Observability stack can quickly become just as complex as the systems it's meant to watch. For most teams, the answer is to pay someone else to deal with it. 17:05 Mountbatten (6th Fl.) Session Observability Are We All on the Same Page? Let’s Fix That - With AI Assistance Luis Mineiro Director of Digital Foundation @ASOS.com, SRE Charmer, Previously @Delivery Hero and @Zalando In distributed systems, incidents rarely fail because of missing signals - they fail because the right people aren’t mobilised quickly enough, and teams struggle to build a shared understanding under pressure.