Agent Harness Engineer - Core Runtime

NinjaTech AISydneyRemoteFull-time3 Sept 2026

NinjaTech AI is seeking a Core Runtime engineer to own the agent harness that powers autonomous AI agents. The role involves designing the run loop, orchestrator, and monitoring layer, and requires 4+ years of Python backend experience and familiarity with LLM APIs.

What you'll do

Design and evolve the agent run loop: context management, turn compaction, and recovery from failed or partial tool calls
Build the orchestrator that launches, supervises, and hands off long-running background agents without losing state
Implement the monitoring layer that detects stuck, looping, or runaway agents and intervenes safely
Own the provider abstraction across model backends (Claude, Codex, and others) so the harness stays model-agnostic
Make the harness observable end to end with structured telemetry, so failures are debuggable in production
Keep the loop fast and cheap at scale while it powers millions of daily user interactions
Research, design, and build the core run loop and orchestration that power autonomous agents
Define how agents manage context, recover from errors, and hand off work cleanly
Build rapid prototypes and proof of concepts to turn harness ideas into shipped product capabilities
Stay up-to-date with agent architectures from research and industry, and fold the best ideas into our runtime
Design self-healing mechanisms so agents degrade gracefully instead of failing hard
Collaborate with cross-functional teams including applied science, product, and integrations

What you need

Must-have

PythonConcurrencyProcess supervisionFault-tolerant designLLM APIsAgentic patternsTool callingFunction calling

Nice-to-have

SandboxingContainerizationSecure code execution

Soft skills

Engineering rigorScientific curiosityCollaboration

Agent HarnessCore RuntimePythonLLMAutonomous AgentsBackendSystems EngineeringRemoteFull-timeAI

/What you'll do

/What you need

Must-have

Nice-to-have

Soft skills

What you'll do

What you need