diff --git a/dev/LOGGING.md b/dev/LOGGING.md
new file mode 100644
index 0000000000..9db2ed8d25
--- /dev/null
+++ b/dev/LOGGING.md
@@ -0,0 +1,71 @@
+# Context: controller-runtime
+
+Since 0.10, Fleet uses [controller-runtime](https://github.com/kubernetes-sigs/controller-runtime) as its Kubernetes
+controller framework. It implements _structured logging_, which basically means logging _constant_ log messages with
+variable key-value pairs.
+
+For instance:
+```
+// Unstructured
+myloglib.Log(fmt.Sprintf("Something happened: %v", err))
+
+
+// Structured
+logger.V(1).Info(
+    "Reconciling bundle, checking targets, calculating changes, building objects",
+    "generation",
+    bundle.Generation,
+    "observedGeneration",
+    bundle.Status.ObservedGeneration,
+)
+```
+
+More info [here](https://github.com/kubernetes-sigs/controller-runtime/blob/main/TMP-LOGGING.md) on how
+`controller-runtime` approaches logging.
+
+Fleet has preexisting logging code using [logrus](https://github.com/sirupsen/logrus), which can be ported to
+[zap](https://pkg.go.dev/go.uber.org/zap), for which `controller-runtime` provides helpers.
+
+# Where to log
+
+Logs should be produced from:
+
+* high-level functions, which do something that the user should know about, such as reconciliation, an entity not being
+found where it was expected, etc. Implementation details, which live in helper functions (eg. resetting a SHA256 sum for
+a manifest), are less relevant for logging.
+
+* functions taking a context as a parameter: these functions can time out, and knowing where the timeout happened before
+  the error bubbles up can help troubleshoot it.
+
+# What (not) to log
+
+At the risk of stating the obvious, no sensitive information should be logged (eg. secrets).
+
+# Levels
+
+Fleet logging makes use of the following levels:
+* `Info` for events about which users may care
+* `Error` for errors, although those could
+[arguably](https://web.archive.org/web/20240521184322/https://dave.cheney.net/2015/11/05/lets-talk-about-logging) be
+logged as `Info` as well.
+
+## Verbosity
+
+Log messages can have a numerical verbosity. The default verbosity is 0, we also use 1 for debug logs and 4 for logs
+that aid in tracing problems.
+If log messages should not be included in the output of a binary started with the `--debug` argument, their level should
+be higher than 5.
+
+# Formatting
+
+Fields used for structured logging must follow camelCase.
+
+# Consistency is key
+
+Check how code living in the same file, or package, or higher up the call stack, does logging. Ensure that constant
+log messages and field names are phrased and formatted in consistent ways.
+
+# Useful links
+
+* [Dave Cheney on
+logging](https://web.archive.org/web/20240521184322/https://dave.cheney.net/2015/11/05/lets-talk-about-logging)
diff --git a/dev/README.md b/dev/README.md
index 122f4356ad..31ad5a6fdf 100644
--- a/dev/README.md
+++ b/dev/README.md
@@ -1,4 +1,8 @@
-# Dev Scripts
+# Logging
+
+Instructions on writing useful, well-formatted log messages can be found [here](./LOGGING.md).
+
+# Testing: Dev Scripts
 
 These scripts are used for running tests locally in k3d. Don't use these on
 production systems.