This works well for vibecoding on a codebase in isolation, which to be fair is what the author is addressing. I don’t think it solves the problems at the current frontier of agent use though, where you expose internal infrastructure via tools to make the agent maximally productive. How to do this safely is still unsolved
Limit access to whatever their project requires. The difference is that human interns have some common sense and won't suddenly be hijacked by a hidden message they stumble upon while searching the web, instructing them to exfiltrate a bunch of proprietary data. It is surprisingly easy to get an agent to do that though
Comments
This works well for vibecoding on a codebase in isolation, which to be fair is what the author is addressing. I don’t think it solves the problems at the current frontier of agent use though, where you expose internal infrastructure via tools to make the agent maximally productive. How to do this safely is still unsolved
I mean this question unironically: how do you do it with interns?
Limit access to whatever their project requires. The difference is that human interns have some common sense and won't suddenly be hijacked by a hidden message they stumble upon while searching the web, instructing them to exfiltrate a bunch of proprietary data. It is surprisingly easy to get an agent to do that though