What would it take for an agent to actually debug a production Android bug? We’ll start with the engineering foundations: structured events, queryable production data, automated instrumentation, and a unified view of product and system telemetry. You’ll also see how newer Android APIs can capture rich Perfetto traces and heap dumps directly from production. Then we’ll put agents to work at increasing levels of autonomy: investigating an issue, proposing a fix, implementing it, validating the result and finding gaps in telemetry to improve what it can see in production. We’ll see real workflows, including the failures and limitations we encountered along the way. You’ll leave with practical patterns for: - Giving agents enough production context and evidence to investigate real bugs. - Using production data as a feedback loop to improve instrumentation. - Designing agent workflows that can take action safely on production issues. - Moving from agent-assisted investigation toward automated bug fixing. - Knowing where automation works well and where human judgment still matters.