AI Harness PM: Architect the Agent Backbone

A65; Singapore; On-site; SGD 120,000 - 180,000; Full time

Benefits offered by this job

Frontier technology

IPO growth opportunity

Job summary

A65 is building a new harness layer between AI models and users, defining the architecture and standards for the agent ecosystem. You will own end-to-end design, from connectors to evaluation pipelines, in a greenfield environment.

You will partner with ML researchers and engineers to ship reliable, safe, and impactful AI-powered advertisers tools, shaping the product roadmap from day one and driving cross-functional collaboration.

Qualifications

• 4+ years of software engineering experience (full-stack or platform)
• Hands-on with testing frameworks, evaluation systems, or developer tooling
• Familiarity with LLM-based systems and AI agents
• Product thinking and ability to translate signals into prioritised decisions
• Strong English communication for cross-functional collaboration

Responsibilities

• Design and own the agent harness framework between model and user
• Build and maintain integrations to internal systems ensuring reliability
• Develop evaluation pipelines and benchmarking for agent performance and safety
• Create feedback loops using real-world tasks to extract product signals
• Collaborate with ML researchers to evolve the harness alongside the model
• Diagnose and troubleshoot agent failures and orchestration issues

Skills

Full-stack engineering

Product thinking

Agile development

CI/CD

English communication

Tools

LangChain

Kubernetes

GitHub Copilot

LangGraph

AutoGen

Job description

A65 is building a new harness layer between AI models and users, defining the architecture and standards for the agent ecosystem. You will own end-to-end design, from connectors to evaluation pipelines, in a greenfield environment.

You will partner with ML researchers and engineers to ship reliable, safe, and impactful AI-powered advertisers tools, shaping the product roadmap from day one and driving cross-functional collaboration.