Design an Autonomous Coding (SWE) Agent

Design an AI agent that takes a GitHub issue, navigates a large codebase, writes a fix, runs tests, and opens a pull request — all without human help.

Code AgentsSandboxingTest-DrivenRepo NavigationVerification

Key Requirements

Navigate and understand a large, unfamiliar codebase
Edit multiple files while keeping changes consistent
Run tests in a sandboxed environment
Avoid reward hacking (e.g., deleting failing tests)
Produce clean, reviewable pull requests

Interviewer Follow-ups

Q1How do you prevent the agent from introducing security vulnerabilities?
Q2How does it handle flaky tests vs. real failures?
Q3How do you retrieve relevant files instead of loading the whole repo?

Design an AI Coding Copilot — Real-Time Inline Completion

Design an Agent Evaluation & Guardrail Platform