AI Software Engineer, Agent Harness

About this job

EnCharge AI is developing a next-generation AI platform featuring a novel in-memory-computing architecture that dramatically enhances compute energy efficiency and performance for AI inference workloads. As an AI Software Engineer focusing on the Agent Harness, you will own the critical layer that executes agents against files, tools, and documents while handling permissions, memory, and unattended execution. Your work involves building and maintaining this architecture using open-source and custom code to serve open-weight models and bespoke checkpoints on EnCharge hardware. A normal day in this role involves designing and building robust agent loops, ensuring reliable tool use across various models, developing internal and customer-facing tools, and creating evaluation harnesses to monitor cost, latency, and quality.

Responsibilities

• Own the end-to-end harness architecture including agent loops, safe execution, context management, and observability.
• Build custom components lacking open-source equivalents such as session semantics and enforced permissions.
• Adapt to frequently changing models by managing prompt formats, stop conditions, and benchmarking.
• Ensure reliable tool usage across different models through validation, retries, and fallbacks.
• Develop and security-review agents, tools, and MCP servers for internal and customer use cases.
• Build evaluation harnesses for regression runs, tracking cost, latency, and quality metrics.

Requirements

• 10+ years of software engineering experience in backend systems or ML infrastructure.
• Strong proficiency in Python and at least one systems language like Go, Rust, or C++.
• Proven experience shipping and operating production agent loops involving multi-step workflows.
• Hands-on expertise with RAG, context management, and memory for LLM applications.
• Experience implementing sandboxing, isolation, and permission models for automated systems.
• Practical experience running open-weight models and understanding serving choices.
• Comfort working in fast-moving, ambiguous environments with strong product instincts.

Good to have

• Contributions to open-source agent harnesses or coding agents.
• Experience with agent benchmarks like SWE-bench and building internal task suites.
• Familiarity with MoE serving such as expert placement and quantization.
• Experience with observability for LLM systems and document parsing pipelines.
• Desktop or GUI application development experience.

Who should apply

This role suits seasoned software engineers with extensive backend or ML infrastructure experience who have successfully shipped production agent loops. Candidates should be comfortable taking ownership of ambiguous roadmaps, defining new architectures, and working closely with cutting-edge AI inference hardware and open-weight models. Freshers are not eligible to apply as the position requires 10+ years of professional software engineering experience.

Frequently asked questions