Anthropic Claude Agent Developer — Managed Agents Harness & Claude Agent SDK

Anthropic Claude Agent Developer - Managed Agents Harness and Claude Agent SDK Ship production AI agents on Anthropic's hosted runtime ================================================ In April 2026, Anthropic launched its Managed Agents harness, a hosted runtime where the agent definition, the sandbox environment, and the session are cleanly decoupled. It is the cleanest production target available for serious Claude agents today. But the surface area is genuinely new: vaults, dreams, outcomes, memory stores, the built-in agent toolset, MCP-proxied tools, custom tools, and session event streams. Our clients want to ship Claude agents that actually run in production: multi-agent coordinators with memory, scheduled dreaming, graded outcomes, end-user OAuth via vaults, and skill bundles attached. Most engineers can spell "Claude API." Very few have actually built and operated this stack. We are expanding our consulting bench and hiring someone who has. ================================================ WHAT YOU WILL DO Architect and build production agents on Claude Managed Agents (api.anthropic.com) and the Claude Agent SDK (self-hosted), picking the right deployment target per client Design multi-agent systems: coordinators, delegate agents, session threading, and event flows Wire up memory stores, dreams, and outcomes for agents that learn and improve across sessions Configure tool surfaces (custom tools, the built-in agent toolset, and MCP toolsets) and build custom MCP servers where needed Implement vault-based authentication and per-end-user OAuth flows for agents accessing third-party systems Manage agent versioning, evals, and the rubric-driven outcomes loop Partner with client engineering, security, and governance teams on identity, audit, and observability ================================================ WHO YOU ARE You have actually shipped agents to production on the Claude Managed Agents harness OR with the Claude Agent SDK. Not demos, not prototypes. Fluent with the canonical primitives: Agents, Environments, Sessions, and Events. You know the managed-agents-2026-04-01 beta header without looking it up. Hands-on experience with at least three of: memory stores at /mnt/memory/, dreams, outcomes and the grader, multi-agent coordinators, vaults and mcp_oauth, the built-in agent toolset (agent_toolset_20260401), MCP server development, Agent Skills bundles Strong Python or TypeScript skills with the Anthropic SDK and the Claude Agent SDK Comfortable with the session event model (session.status_idle, requires_action, user.tool_confirmation, user.custom_tool_result) and building reliable event consumers Excellent written and verbal communication; able to brief client executives and partner with their engineering teams ================================================ ALSO A PLUS Production experience with prompt caching strategies specific to agent loops MCP server development beyond toy examples: auth, rate limiting, observability Past work on agent evals, graders, or rubric design Contributions to anthropic-sdk-python or claude-agent-sdk-python/typescript, or filed reproducible bug reports against them ================================================ WHAT TO INCLUDE IN YOUR PROPOSAL A SHORT VIDEO (2 TO 3 MINUTES) INTRODUCING YOURSELF AND WALKING THROUGH YOUR EXPERIENCE. THIS IS REQUIRED. PROPOSALS WITHOUT A VIDEO WILL NOT BE REVIEWED. One specific example of a Claude agent you have built and shipped: Managed Agents or Claude Agent SDK, the use case, the architecture (which primitives, which tools, which memory/outcomes setup), and the outcome Which parts of the Anthropic agent stack you use today, and how A short note on your background: years building agents on Claude, scale of past deployments, and any open-source contributions or repos we can review Your availability ================================================ If you have already shipped on the Anthropic harness and want to help enterprises do the same, we want to talk.