Author(s): Eivind Kjosbakken

Originally published on Towards AI.

How to Safely Run Coding Agents

Coding agents such as Claude Code and Codex have provided me the biggest efficiency boost I’ve ever experienced while programming, way more of a boost compared to getting more powerful computers or learning new topics and techniques.

After the intro, the author explains why coding agents are worth using despite common concerns: in many cases they write functional code equal to or better than humans and can discover bugs effectively—especially when agents are used to review each other’s work. They also argue that “unsafe actions” aren’t the core problem if permissions and infrastructure are designed properly; irreversible mistakes should be made hard or impossible through permission scoping and backups. The article then provides a practical approach to permissions, typically being liberal with what agents can do but skipping permission prompts (YOLO/dangerously-skip-permissions) except for clearly destructive commands like recursive delete (rm), which should require explicit approval. Finally, it shares techniques for safer operation, including restricting irreversible commands, using an additional agent for iterative code review, and adjusting the level of care based on whether the work is in high-security domains.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI