Agent Harness Engineer

必須

• Bachelor's degree or equivalent practical experience in Computer Science, Software Engineering, Artificial Intelligence, Machine Learning, Mathematics, Physics, or related fields
• 5+ years of practical experience as a backend engineer
• Production product development experience in Python
• Experience designing and implementing production systems that leverage LLM / AI agents
• Experience designing and implementing distributed systems (including design and coding, not just operations)
• Experience designing and implementing RESTful APIs / gRPC
• Japanese: Fluent — able to discuss product development without friction
• English: Business level

歓迎

• Agent Framework / Agent Harness design and implementation experience (LangChain / LangGraph / AutoGen, etc.)
• Production operations experience on cloud platforms (AWS / GCP / Azure)
• Understanding of RAG systems, vector databases, and memory architectures
• Model routing and inference optimization experience
• Foundation software development experience in Go (SDKs, runtimes, frameworks, etc.)
• Deep understanding of Kubernetes / container orchestration
• Event-driven architecture experience (Kafka / RabbitMQ, etc.)
• Experience implementing safety guardrails, policy execution, and AI observability
• ML infrastructure / MLOps construction experience
• Technical communication ability in English

使用ツール

Why We're Hiring

• 2025 was "the year of AI agents." 2026 is _"the year of Agent Harness."_
• In a world where JAPAN AI STUDIO autonomously executes hundreds of workflows as "the brain of the enterprise," agent performance is not determined by the model alone. The Agent Harness — the control layer that wraps the model and manages session state, checkpoints, guardrails, context injection, and tool execution — is the key that transforms an agent from "works in a demo" to "trusted in production."
• "The brain of the enterprise" approves requests, allocates resources, and discovers prospects — the Agent Harness is the heart that controls each of these actions safely, quickly, and reliably.
• JAPAN AI is hiring Agent Harness Engineers to design and implement this Agent Harness in-house and build it as the shared foundation across all products.

Mission

• _"Design the heart of 'the brain of the enterprise.'"_
• Design and implement the Agent Harness — execution engine, orchestration, guardrails, memory, and model routing — that enables AI agents to operate safely, quickly, and reliably. Build the control foundation for hundreds of workflows running on JAPAN AI STUDIO, entirely in-house.

What Is an Agent Harness?

• An Agent Harness is the control and execution infrastructure layer that wraps AI models. While Agent Frameworks (e.g., LangChain) handle agent _construction_ , the Agent Harness handles agent _control and operation_ .
• Backend Engineer
• What you build : Web APIs, microservices
• Relationship with AI/ML : Calls ML models via API
• State management : Stateless request/response
• Safety controls : Authentication, authorization, input validation
• Agent Harness Engineer
• What you build : LLM-centric agent execution engines, SDKs, orchestrators
• Relationship with AI/ML : Designs model routing, RAG integration, context injection, and inference optimization at the system level
• State management : Agent session management, checkpoints, long-term memory, working memory
• Safety controls : Guardrail/policy execution engine — a rule execution layer that controls LLM output

Role & Expectations

• As an Agent Harness Engineer, you will design and implement the agent control and execution infrastructure, leveraging your AI/ML knowledge.
• Design and implement the execution engine (Graph Runtime / State Machine) with deep understanding of LLM / AI agent operating principles
• Own AI-specific system design including model routing, context management, and memory infrastructure (long-term memory, working memory)
• Design and develop the Agent SDK used by 120 in-house engineers
• Build the guardrail / policy execution engine to safely control agent behavior
• Collaborate with Research Engineers to integrate the latest research outcomes into the production infrastructure

Why You'll Love This Role

• _Build the Agent Harness in-house_ — Design and implement the hottest architectural concept of 2026 without relying on OSS. Stand at the industry's cutting edge.
• _At the intersection of AI/ML × Backend_ — Design and implement the agent execution infrastructure with deep understanding of LLM operating principles. Neither pure infrastructure nor pure ML — a new domain.
• _Foundation software designer_ — This is not a job writing YAML. You will build SDKs, execution engines, and orchestrators in code. Low-level knowledge directly applies.
• _Developer experience architect_ — Design the SDK and toolchain used by 120 in-house engineers, improving productivity across the entire development organization.
• _Powering every product_ — In a production environment used by ~200 companies, every AI agent runs on the Harness you build.
• _Rapid-growth environment_ — In a startup that has grown to 200+ people and 9 products in just 3 years, you will have significant autonomy in technical decision-making.

Agent Harness design & implementation

• Design and implement the agent execution engine (Graph Runtime / State Machine)
• Design and develop the Agent SDK — the interface for in-house engineers to build agents
• Implement session management, checkpoint, and recovery mechanisms
• Build the guardrail / policy execution engine — a rule execution infrastructure that controls agent behavior

AI/ML System Integration

• Model routing — optimal routing of inference requests across multiple LLM providers and model types
• Design context management and memory infrastructure (long-term memory, working memory, RAG integration)
• Optimize inference pipelines (latency reduction, cost efficiency, caching strategies)
• Integrate latest research findings into the production infrastructure in collaboration with Research Engineers

Orchestration & performance

• Develop workflow orchestration and queuing systems
• Cost/performance optimization (autoscaling, caching, batch processing)
• Inference request routing and load balancing

Reliability & Operations

• Maintain platform uptime of ≥99.9%
• Incident response and post-mortems
• Design data access and permission management infrastructure

Key Results (KRs / Metrics)

• Agent SDK adoption rate (in-house team usage rate and satisfaction)
• Agent execution success rate (task completion rate, checkpoint recovery success rate)
• Harness-attributed failure rate (guardrail breach rate, state inconsistency rate)
• Execution latency P95 / P99 (Harness layer overhead)
• Inference cost efficiency (cost optimization through model routing)
• Developer experience score (internal NPS for SDK / API)

Team Structure

• Approximately 120 members are part of the development organization.
• Agent Harness Engineers work across the following groups:
• Infra — Cloud infrastructure and SRE
• Data — Data pipelines and analytics infrastructure
• Agent Harness — Agent execution framework
• Closely collaborating roles:
• Agentic Product Engineer — Agent feature development (SDK users)
• Research Engineer — R&D and integration of new methods into the infrastructure
• AI Quality Scientist — Evaluation pipeline collaboration
• Product Manager — Product design and non-functional requirements definition

Tech Stack

• Languages : Python, Go (backend / infrastructure), TypeScript / React / Next.js (frontend), NX
• Infrastructure : GCP (containers / K8s), Docker, Terraform
• Messaging : Kafka, Pub/Sub
• Monitoring : Prometheus, Grafana, OpenTelemetry
• Tools : Slack, Confluence, Linear, Google Workspace, GitHub, Notion
• AI Dev Support: Claude Code MAX Plan, Cursor, ChatGPT, Devin
• Workstation : Mac (Apple Silicon), dual monitor setup