AI Agent Engineer / Agentic Systems Engineer
Open a video or resource link, watch it, then check its box to track your progress and unlock what comes next.
- Build the agent harness and manage the context it reasons over
- Connect agents to tools and orchestrate secure multi-step systems
- Operate agents with eval-driven reliability, grounding and economics
Benchmark: Zoom Principal Agentic AI Engineer + OpenAI AI Systems Engineer, Codex Agents (current Sep 2026) Market refresh: agent harnesses, orchestration, tool use, MCP, context and memory, sandboxing, multi-agent coordination, evaluations, latency/cost, observability, production reliability and full-lifecycle ownership of agent infrastructure used by downstream products.
Who’s Hiring Right Now
US openings posted in the last 7 days at OpenAI — this role’s own benchmark employers — refreshed daily, as of 2026-11-01. Pulled directly from each company’s public job board; not exhaustive, and postings close fast — worth confirming the details on the employer’s own site before applying.
Build the agent harness and manage the context it reasons over
Agent architecture, reasoning loops + harness fundamentals
Coverage: AI Agents / Agentic Systems / Software Engineering
Context engineering, memory + retrieval
Coverage: Context Engineering / RAG / Agent Memory
Connect agents to tools and orchestrate secure multi-step systems
Tool use, MCP + integration contracts
Coverage: MCPs / APIs / Agentic Tools Platform / Security
Multi-agent orchestration + platform architecture
Coverage: Agent Orchestration / Distributed Systems / AI Platform
Operate agents with eval-driven reliability, grounding and economics
Agent evals, observability + production reliability
Coverage: AI Evals / LLMOps / Observability / SRE
Grounding, latency/cost + production agent operations
Coverage: RAG / AI Platform / Reliability / Cost Optimization
Role-Readiness Proof / Capstone
Ship a production-grade agent system for a real multi-step workflow. Implement a Python/TypeScript agent harness with tools and MCP; use context engineering, memory and grounded retrieval; add sandboxing and least-privilege tool permissions; orchestrate a specialist/sub-agent path; build eval sets for reasoning and tool success; instrument traces, latency, cost and failure categories; deploy with SLOs, rollback and a runbook; then iterate from measured production-style failures.
Proof Rubric
- Prove task completion, not a demo: robust harness design
- correct tool/MCP contracts
- context and memory discipline
- grounded retrieval
- safe execution boundaries
- multi-step/multi-agent coordination
- eval coverage and regression gates
- observability and diagnosable traces
- latency/cost budgets
- production resilience and incident handling
- measurable improvement from failures
- and clear tradeoffs between autonomy, reliability, cost and human control