Atomic

A new open source software project called Atomic aims to make artificial intelligence coding tools more predictable. Developed by Bastani Inc and hosted on GitHub, the project acts as a control system for managing AI programming assistants. Rather than letting an AI write code without supervision, developers can outline a step-by-step process using everyday language, supported by automated testing loops and required human approval checkpoints.

This approach tackles a growing headache in software engineering. While AI can draft code rapidly, it frequently introduces subtle bugs or invalid instructions that break existing applications. By embedding strict verification checks directly into the creation pipeline, development teams can catch errors early and prevent unverified AI output from reaching live software.

For everyday computer users, this movement toward verifiable AI development could lead to more reliable apps and fewer broken updates. However, key questions remain about how well natural language guardrails can handle complex edge cases. It is also unclear whether adding multiple human and automated approval steps will eliminate the speed advantages that AI coding tools promised in the first place.