Skip to content

Running agents as a team

The required reading is the notes and the studio guide; the week's row in the syllabus calendar lists what is due. Before the first session, read the learning objectives and the event table in the continuing dashboard case, and write which observation separates a failed request from a lost reply.

The four sources below are optional. Each entry names the part to read and what to use it for.

  • Budget and logging boundaries (about 12 minutes): read Virtual Keys — LiteLLM, especially “Upperbound /key/generate params”, and Logging — LiteLLM, especially “Redact Messages, Response Content”. Compare those controls with the gateway jobs in the notes (key held outside the work, listed routes, per-run caps, minimal logs), then say which checks would still need a record to show the setting enforced. (tools/litellm; as_of: 2026-09-05; Source 1; Source 2)
  • A postmortem is reviewed before it counts (about 20 minutes): read the opening and the “Google’s Postmortem Philosophy” and “Collaborate and Share Knowledge” sections of Postmortem Culture: Learning from Failure, Chapter 15 of Google's Site Reliability Engineering book. It gives the parts a postmortem must cover, the triggers that call for one, a blameless way to describe contributing causes, and review questions such as whether the root cause is deep enough. Use those questions on the draft root-cause analysis in the notes. (studies/sre-postmortem-culture; as_of: 2026-10-09; Source 1)
  • Alerts, rollback and changes (about 15 minutes): read the “Monitoring”, “Emergency Response” and “Change Management” sections of Introduction, Chapter 1 of the same book. It sorts monitoring output into alerts, tickets and logs, and treats changes to a live system as a common cause of outages, to be rolled out gradually and rolled back safely. Use it to decide which signals of your own project should interrupt someone and which only belong in a log, and to check that a rollback ends with a runnable check of the user path. (studies/sre-introduction; as_of: 2026-10-09; Source 1)
  • Telemetry is not a storage backend (about 12 minutes): read What is OpenTelemetry? and Collector — OpenTelemetry, focusing on the signals and the collector's role. Use them to choose the smallest trace, metric and log fields an incident like this week's would need, and to state what a collector failure would leave unknown. (tools/opentelemetry; as_of: 2026-09-05; Source 1; Source 2)

For each source you read, write one sentence that connects its claim to a line of your studio note and one sentence naming what the source does not show about your own project. Do not report a source's example as your own result.