References: Subagents and Claude Managed Agents

Source material

Source curriculum (structural mirror, cited as further study):
• Anthropic Academy (https://anthropic.skilljar.com/):
    "Introduction to Subagents" + the Agent SDK + Claude Code 101 /
    Claude Code in Action courses
  License: Anthropic Academy course content is account-gated;
    Clawdemy structurally mirrors the Academy's lesson progression
    as inspiration and cites it as further study. Every substantive
    claim in this lesson is verifiable against the public Anthropic
    documentation.

Primary public-doc anchors (every substantive claim verified
against):
• Anthropic, "Subagents in the SDK" (Claude Agent SDK):
   the verbatim definition, four benefits, three creation paths,
   AgentDefinition fields, inheritance rules, parent-to-subagent
   channel, and Python + TypeScript examples
  https://code.claude.com/docs/en/agent-sdk/subagents
• Anthropic, "Claude Managed Agents overview"
   (the verbatim positioning, four core concepts, five-step flow,
   built-in tool set, data-retention posture)
  https://platform.claude.com/docs/en/managed-agents/overview
• Anthropic, "Get started with Claude Managed Agents"
   (the three endpoints POST /v1/agents + /v1/environments +
   /v1/sessions, the event-stream shape, code examples in eight
   languages, the agent_toolset_20260401 toolset)
  https://platform.claude.com/docs/en/managed-agents/quickstart
• Anthropic, "Claude Code subagents" (the filesystem-based
   subagent definition path used inside Claude Code)
  https://code.claude.com/docs/en/sub-agents
• Anthropic, "Dynamic workflows" (the Workflow tool for
   coordinating dozens to hundreds of subagents outside
   the per-turn context)
  https://code.claude.com/docs/en/workflows
• Anthropic engineering, "Building Effective AI Agents"
   (the L9 patterns this lesson realizes; cross-referenced)
  https://www.anthropic.com/engineering/building-effective-agents

Verbatim claims sourced from the public docs:
• Subagent definition: "Subagents are separate agent instances
   that your main agent can spawn to handle focused subtasks.
   Use subagents to isolate context for focused subtasks, run
   multiple analyses in parallel, and apply specialized
   instructions without bloating the main agent's prompt"
   (Subagents in the SDK)
• Context isolation: "Each subagent runs in its own fresh
   conversation. Intermediate tool calls and results stay inside
   the subagent; only its final message returns to the parent"
   (Subagents in the SDK)
• Channel constraint: "The only channel from parent to subagent
   is the Agent tool's prompt string, so include any file paths,
   error messages, or decisions the subagent needs directly in
   that prompt" (Subagents in the SDK)
• Managed Agents positioning: "Pre-built, configurable agent
   harness that runs in managed infrastructure. Best for
   long-running tasks and asynchronous work" (Managed Agents
   overview)
• Managed Agents value: "Claude Managed Agents provides the
   harness and infrastructure for running Claude as an autonomous
   agent. Instead of building your own agent loop, tool execution,
   and runtime, you get a fully managed environment where Claude
   can read files, run commands, browse the web, and execute code
   securely" (Managed Agents overview)
• Beta header: "All Managed Agents endpoints require the
   managed-agents-2026-04-01 beta header. The SDK sets the beta
   header automatically" (Managed Agents overview)
• Data-retention posture: "Claude Managed Agents is stateful by
   design ... not currently eligible for Zero Data Retention or
   HIPAA Business Associate Agreement (BAA) coverage" (Managed
   Agents overview)

Required attribution: "Based on the structure of the Anthropic
  Academy 'Introduction to Subagents' + 'Building with the Claude
  API' courses (https://anthropic.skilljar.com/) and Anthropic
  engineering, 'Building Effective AI Agents' (Erik S. and Barry
  Zhang, 2024-12-19). This lesson is an independent structural
  mirror in original prose; every substantive claim is verified
  against the public Anthropic Claude Agent SDK documentation at
  https://code.claude.com/docs/en/agent-sdk/ and the public
  Claude Managed Agents documentation at
  https://platform.claude.com/docs/en/managed-agents/.
  Anthropic does not endorse it."

Going deeper

A short, durable list. Each link is a specific next step inside Track 22.

Lesson 12 of this track, “Shipping a Claude application.” Where whichever primitive you chose here (self-built L8 loop, Subagents fan-out, or Managed Agents) goes to production: cost monitoring on usage + usage.iterations, eval-set discipline, the rollout checklist, the per-feature data-retention check.
Track 20 (AI Agents and Tool Use): the full track-level treatment of agent design, harness engineering, and operational discipline. T22 L11 here is the Anthropic-specific instance; T20 covers the broader design principles.

Adjacent tracks (the natural next destinations)

Track 20 (AI Agents and Tool Use): pick this if you want full track-level depth on agent design and harness engineering. The Subagents + Managed Agents primitives this lesson covers fit inside the broader treatment T20 develops over ten lessons.
Track 21 (LLM Ops and Production): pick this if you want the provider-agnostic view of operational discipline for agents. The eval-set discipline this lesson defers to L12 is from T21 L7 “LLMOps.”

Where this connects inside the track

The two primitives the rest of T22 builds toward:

Lesson 1 (first call): the messages.create call is what every Subagent and every Managed Agents session ultimately runs internally. The iterate-the-content-array discipline applies to the events Managed Agents streams back.
Lesson 3 (model selection): the per-Subagent model field is L3’s mix-and-match applied per-step. The effort dial on AgentDefinition.effort is L3’s effort parameter.
Lesson 4 (custom client tools): Subagents inherit (or restrict) the parent’s tool set from L4.
Lesson 5 (server tools + Anthropic-schema clients): Managed Agents’ agent_toolset_20260401 is the productionized version of L5’s bash, text_editor, web search, web fetch, plus file operations.
Lesson 6 (MCP): Subagents take mcpServers in AgentDefinition; Managed Agents takes MCP servers in the agent definition. Same MCP from L6.
Lesson 7 (prompt caching + context management): Managed Agents includes built-in prompt caching and compaction (the docs say so explicitly). Subagent context isolation means the parent does not pay for the subagent’s intermediate work, which is a complementary cost lever.
Lesson 8 (agent loop): Subagents run inner L8-style loops. Managed Agents IS an L8-style loop, Anthropic-hosted and productionized.
Lesson 9 (six patterns): Subagents realize patterns 3 (parallelization), 4 (orchestrator-workers), and 6 (autonomous agent) directly. Managed Agents fits pattern 6 with the harness lifted.
Lesson 10 (Agent Skills + Claude Code): Subagents take skills in AgentDefinition (L10’s preload pattern, per-subagent). Claude Code’s filesystem-based agents at .claude/agents/ are the filesystem variant of this lesson’s programmatic AgentDefinition.
Lesson 12 (next, shipping): cost monitoring, eval-set, rollout checklist, data-retention posture per primitive.

References: Subagents and Claude Managed Agents

Source material

Read this next

Going deeper

Adjacent tracks (the natural next destinations)

Where this connects inside the track