Staff Engineer, Harness Engineering

Harper•San Francisco, CA

1d•$253,000 - $308,000•Onsite

About The Position

Harper is building an AI-led commercial insurance distribution platform, aiming to serve the 36 million businesses in America that require insurance, addressing the significant underinsurance problem. The company is experiencing rapid growth, adding approximately 1,000 customers per month and growing 100x year-over-year, and is scaling towards a Series B funding round. The core of Harper's technology is 'Harper agents,' which include product agents (handling intake, sales, service, voice) and coding agents (frontier coding agents, OpenClaw, Hermes-based agents). All these agents operate on a shared substrate called the 'harness.' This role involves owning and developing the meta-harness, the foundational layer upon which all Harper agents are built. The company's thesis is that success with agents comes from a robust harness that enables models to function as reliable workers, encompassing the loop, tools, memory, execution environment, budgeting, fallback paths, and guardrails. The harness is seen as the company-specific leverage, making it crucial for all agent types, including product, coding, voice, underwriting, and service agents. A successful harness will accelerate agent development, reduce failure rates, and facilitate learning from previous systems.

Requirements

8+ years software engineering experience, including senior+ scope at a high-growth company.
Production agent-harness or AI-substrate experience, covering agent loop, tool routing, execution environment, and model routing.
Strong written communication skills, including experience with RFCs, API contracts, and integration guides.
Based in San Francisco or willing to relocate.

Nice To Haves

Open-source contribution to agent/harness frameworks.
Depth in sandbox/isolation infrastructure.
Experience with foundation-model partners or early-access programs.

Responsibilities

Own the meta-harness, the core infrastructure for all Harper agents.
Develop and maintain the agent loop, including prompt construction, tool routing, context-window management, and retry/timeout/budget logic.
Build and manage the execution environment, including sandbox lifecycle, isolation, blast-radius limits, and file-system/network policies.
Define and manage the tool layer, including schema, authentication, rate limiting, and observability for agent tools.
Develop and maintain the model-provider abstraction, handling provider routing, fallback chains, cost/latency tradeoffs, and eval-driven model selection.
Implement multi-agent coordination patterns such as parent/child handoff, shared memory contracts, parallel execution, conflict resolution, and subagent budgets.
Create and maintain the harness SDK for pod engineers.
Implement guardrails, including banned tool combinations, prompt-injection defense, data-egress policies, and PII scrubbing.