Toil is the repetitive operational work that keeps a system running but never improves it, and reducing it is the core purpose of SRE.
Toil, in Site Reliability Engineering, is the manual, repetitive, automatable operational work that scales linearly with a service and provides no lasting value, the kind of work that keeps a system running but never makes it better.
Google's SRE practice defines toil precisely: work that is manual, repetitive, automatable, tactical rather than strategic, without enduring value, and that scales with the size or load of the service. Restarting a stuck service by hand, applying the same fix to the same recurring problem, running a routine task manually each week, these are toil. It is worth distinguishing toil from overhead, the meetings and administration that are not directly operational, and from genuine engineering, which produces lasting improvement. Toil is specifically the operational busywork that a machine could do.
Toil matters because it consumes exactly the capacity that would otherwise reduce future incidents. Every hour spent hand-running an operational task is an hour not spent automating it away or fixing the root cause, which makes toil self-perpetuating: the more time a team spends on it, the less time it has to eliminate it. Toil also burns people out, because repetitive, low-value work is demoralising. This is why SRE caps toil, commonly aiming to keep it under half of an engineer's time, so there is always capacity left for the engineering that makes toil shrink.
Reducing toil follows a clear sequence: measure it, so you know where it is and how much; automate the repetitive tasks that a machine can do reliably; and eliminate root causes so the toil-generating problem stops recurring at all. Increasingly, the frontier is letting autonomous systems handle the well-understood operational work entirely, so the common incidents that generate the most toil are resolved without a human. Reducing toil is not a side project in SRE; it is the point of it.
Ops Singularity eliminates toil at its source: Sentinel AI resolves the common, well-understood incidents autonomously through governed Action Tickets, removing exactly the repetitive operational work that SRE spends its energy trying to automate away, and freeing engineers for the work that reduces future incidents.
Manual, repetitive, automatable operational work that scales with a service and produces no lasting value, keeping the system running without improving it. Reducing it is the core goal of SRE.
It consumes the capacity that would otherwise prevent future incidents, is self-perpetuating, and burns people out. SRE caps toil so engineering time is protected.
Measure it, automate the repetitive tasks, eliminate the root causes, and increasingly let autonomous systems resolve the well-understood incidents that generate the most toil.
Ops Singularity turns open telemetry into autonomous, governed resolution. See it on your own stack.