AI Harness Engineer, Agentic & Autonomous Operations

About the Role

Build the software systems that turn language models into reliable agents for operational work. Working closely with engineering and leadership, you will shape agent architecture from design through production operations.

What You'll Do

• Design orchestration, context assembly, and execution layers for reliable agents.

• Build systems for task decomposition, state management, model selection, execution loops, checkpoints, retries, and human handoffs.

• Connect agents to APIs and operational systems using explicit tool contracts, permissions, validation, and outcome verification.

• Develop retrieval, memory, and knowledge-maintenance strategies that balance relevance, latency, and cost.

• Create evaluation suites and automated tests for decision quality, tool use, task completion, safety, and failure patterns.

• Make agent behavior inspectable with structured traces, state, outcomes, and error classifications.

• Use evaluations, production traces, and controlled experiments to improve prompts, context, tools, and architecture.

• Ship and operate production Python software, investigating incidents and making architectural decisions with the engineering team.

What We're Looking For

• At least 2 years of professional software engineering experience and strong production software engineering fundamentals.

• Experience building, testing, and operating reliable production systems in Python.

• Substantial experience building LLM-powered automation, workflow, or decision systems beyond notebooks, chat interfaces, or basic retrieval pipelines.

• Practical knowledge of LLM orchestration, state management, context assembly, structured outputs, permissions, retries, fallbacks, observability, and evaluations.

• Experience connecting models to real work through relevant evidence, bounded decisions, tool or API use, and verification of outcomes.

• Sound judgment designing for uncertainty and partial failure, including when systems should act, retry, stop, or escalate.

• A systematic approach to evaluation using representative cases, traces, outcome checks, and regression tests.

• Experience shipping and supporting software used by people or operational processes, including debugging timeouts, inconsistent data, changing APIs, and production incidents.

• Workflow engines, distributed systems, developer tools, automation platforms, or integration-heavy backend experience is valuable.

Location

On-site in Berlin, Germany. Visa sponsorship is not available.