Sub-agent orchestration and observability with Claude
Code: ENG-002 | Track: technical | Duration: 0.5 day
Custom sub-agents, orchestration patterns, hooks as the control plane, and telemetry-based observability for Claude Code.
Context
Teams running multi-agent work who need to see what their agents actually did.
Who it's for
Engineering and platform teams
Prerequisites
- Claude Code with telemetry enabled; access to an OTel-compatible backend for the dashboard exercise
Format
- Remote
Agenda
- The agent loop and why sub-agents exist - context windows as the binding constraint, and when isolated sub-agent contexts beat a single thread.
- Building custom sub-agents - hands-on construction of scoped sub-agents with role-appropriate tool access.
- Orchestration patterns and trade-offs - fan-out, pipelines and orchestrator-worker patterns, with the token economics of each.
- Hooks as the control plane - gating risky actions, enforcing guardrails and emitting custom signals through lifecycle hooks.
- Observability with telemetry - hands-on wiring of live sessions to dashboards, then tracing and debugging a misbehaving multi-agent run.
- Governance and production rollout - runtime controls and telemetry as evidence for what your agents did and why.
Session details
- Total time: 0.5 day (4 hours including two short breaks)
- Participants: Up to 15
- Delivery: Remote or in person (UK; other locations by arrangement)
- What you leave with: A scoped sub-agent library, hook configs and a live telemetry dashboard