Software Engineer, Agent Harness
Build the agent harness that plans tasks, calls tools, uses memory, recovers from failures, and operates reliably at scale. The role requires strong reasoning about LLM behavior and the ability to translate it into robust infrastructure.
About the job
Responsibilities
- Build the agent harness that powers workflow and table creation, knowledge bases, and everyday tasks.
- Design and implement how the agent plans, calls tools, uses memory, recovers from failure, and runs reliably at scale.
- Support infrastructure for more than 100,000 builders.
Requirements
- Ability to reason clearly about large language model behavior, including tool calls, context, retries, and evaluations.
- Ability to turn that understanding into robust infrastructure.
Compensation
- $210,000–$260,000 base salary.
- 0.2%–0.8% equity.
Skills
LLMs, Tool Calls, Context Management, Retries, Evaluations, Infrastructure, Memory Systems
Similar jobs
Fullstack Engineering jobs
Sim
San Francisco, CA
Build and own core product capabilities for an AI workspace, spanning usability, reliability, performance, and user experience from concept through production. The role requires strong product engineering, an interest in AI models and design, and end-to-end ownership.
Recall
San Francisco, CA
Build and own internal applications, integrations, and automations that improve sales, finance, billing, and company-wide workflows. The role requires strong product judgment, cross-functional communication, and hands-on experience shipping applications, APIs, and SQL integrations.
Descript
San Francisco, CA
Build and own customer-facing product features across Descript’s editor, media, growth, monetization, and AI surfaces. The role requires 5+ years of full-stack product engineering experience, strong product judgment, and expertise with modern web technologies and experimentation.
Perplexity
San Francisco, CABuild and ship high-quality desktop and Comet applications at Perplexity AI. Requires 4+ years experience shipping native apps, strong architecture skills, and the ability to own projects across Rust, Kotlin, Swift, TypeScript and AI stacks.
Kaizen Labs
Washington, DCBuild and lead full-stack, AI-powered software for government services using the Palantir stack and modern web technologies. The role requires 5+ years of engineering experience, active Top Secret or TS/SCI clearance, and the ability to work hybrid from New York City or Washington, DC.