Software Engineer, AI-Driven Coding Agents

Google LLC; San Jose (CA); On-site; USD 147,000 - 210,000; Full time

Job summary

Google LLC in San Jose, CA, is hiring a Software Engineer for the Self-Maintaining Codebase (SMC) team to advance the intelligence, reliability, and autonomy of our coding agents.

You will build data and evaluation flywheels for large-scale agentic software engineering, driving automated skill improvement and hillclimbing success across multi-repo tasks, including automated refactoring and proto optimization.

Qualifications

• Bachelor’s degree or equivalent practical experience.
• 2 years of Python coding experience.
• 2 years of experience with Generative AI.
• 2 years of experience with Large Language Models.
• Familiarity with Google’s AI-driven coding tools.

Responsibilities

• Work on the core feedback loop that drives the intelligence, reliability, and autonomy of our next-generation coding agents.
• Build data and evaluation flywheel for large-scale agentic software engineering: transforming execution trajectories into actionable loss analysis.
• Pioneer automated skill improvement workflows, and hillclimbing agent success rates across complex, multi-repo codebase tasks.
• Focus areas include AI verifier, quality hillclimbing, agent skill auto-improvement, coding agent evaluation, trajectory analysis.
• Note this role will not focus on agent infrastructure.

Skills

Python

Generative AI

LLMs

Education

Bachelor's degree or equivalent practical experience

Master's degree or PhD in Computer Science or related technical fields

Tools

Agentic AI

Evaluation methodologies

Prompt engineering

Google AI coding tools

Code maintenance & migration

Job description

Google LLC in San Jose, CA, is hiring a Software Engineer for the Self-Maintaining Codebase (SMC) team to advance the intelligence, reliability, and autonomy of our coding agents.

You will build data and evaluation flywheels for large-scale agentic software engineering, driving automated skill improvement and hillclimbing success across multi-repo tasks, including automated refactoring and proto optimization.