SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Netskope is seeking an AI Agent Developer to build production-grade agentic systems that move beyond playground demos into secure, scalable, auditable deployments. This role focuses on engineering discipline in AI agent development—not prompt tweaking, but spec-driven architecture.
You will design and build orchestrator agents that coordinate work across specialized sub-agents, establish comprehensive test cases, and integrate tool gateways. You'll work closely with Product, Security, and Cloud Architecture teams to ensure every agent operates within designated guardrails and risk tiers.
Key responsibilities include:
- Writing full agent specifications (purpose, capabilities, input/output contracts, guardrails, tool boundaries) with safety-critical rigor
- Designing orchestrator architectures that coordinate work without performing it directly
- Developing and testing system prompts and tool schemas that remain accurate across 15–20+ tools
- Classifying agent data sensitivity and setting risk tiers in collaboration with Data Stewards and Product Architects
- Deciding whether each capability should use deterministic logic or model judgment, with corresponding test coverage
- Building agents against platform tool gateways to ensure all actions stay within registered permissions
- Designing for scale: isolated sessions, predictable failures, no single-test-run assumptions
- Preparing production-ready evidence: test results, guardrail coverage, behavioral validation
Required experience: 4+ years building production software, with 1–2 years specifically on LLM-powered agents (not chatbots or RAG). Hands-on expertise with agent orchestration frameworks (LangGraph, Bedrock AgentCore, CrewAI, AutoGen, or similar). Working knowledge of Model Context Protocol (MCP) for server/client integration. Disciplined prompt engineering practice with versioning and test-case validation. Ability to write precise specifications with clear boundaries and failure modes. AWS familiarity (Lambda, IAM, Bedrock) to understand agent runtime behavior.
