Agentic Software Engineer (Harness Engineering)

Quick Facts

• Focus: building and operating a production-grade, AI-first development system using software agents
• Role: Agentic Software Engineer (Harness Engineering)
• Core backend context: PHP

Description

You will develop a production system that uses AI coding agents, with full ownership of the code those agents produce. The work focuses on designing agent harnesses and workflows, splitting tasks across multiple agents, and implementing gates, checkpoints, feedback loops, and automated verification. You’ll also control context/specification/architecture drift and manage agent cost (tokens, iterations, context size, model choice).

Responsibilities

• Develop a production AI-agent-based system for software delivery
• Design harnesses and workflows for AI agents
• Delegate work among agents and manage autonomy, context, and verification
• Build gates, checkpoints, feedback loops, and automated verification
• Prevent and detect drift and ensure code consistency across multiple agents
• Decide when verification can be handled by tests/static analysis vs. when human-in-the-loop is required
• Monitor and reduce agent operating cost (token usage, iteration count, context size, model selection)
• Improve the agent execution environment to enable safe, economical scaling of delegated development

Requirements

• Practical experience with agentic engineering and tools such as Claude Code, Codex, Cursor, or similar
• Strong software engineering fundamentals: architecture, testing, Git, CI/CD, and experience with large existing codebases
• Ability to work effectively in a PHP backend codebase
• Ability to explain and apply the principles of agentic coding vs. “vibe coding,” including well-designed agent harnesses
• Experience designing gates, verification loops, and methods to detect hallucinations and drift
• Experience managing context and multiple agents on larger tasks
• Experience managing agent economics: measuring/limiting token burn, controlling iteration cost, selecting appropriate models, and spotting unproductive context burning
• Ability to ensure agents not only generate code but also verify solutions without breaking other parts of the system
• You do not need to write most code manually; you must demonstrate effective delegation with agents and safe control of the resulting changes

Benefits

Not specified in the provided posting.