Have you ever let an AI agent run a command on your production server, telling yourself it would be fine just this once?
In the latest episode of the No Compromises podcast, Aaron and I discuss why giving AI agents access to production is as reckless as the bad old days of FTPing code straight to the server. An agent is non-deterministic and carries no responsibility, which makes it far more dangerous than a developer making a manual mistake.
We also get into why you shouldn't blindly run complex chained commands an agent suggests, how to verify what your agent is actually doing, and why a little friction in your workflow is worth the safety it buys.
- 00:00 FTPing to production is still wrong in 2026
- 01:50 Running Tinker on production is no better
- 02:28 AI agents on production servers is even worse
- 07:11 How to verify what your agent is actually doing
- 11:03 Silly bit
And after listening, don't forget to subscribe to the podcast, so you don't miss future episodes.
Here to help,
Joel
P.S. We use AI every day, but with the same care we've always brought to our work. If you want help from developers who take responsibility for your project, see how you can work with Aaron and me.
