How has your coding agent setup evolved over the time?
Heat trend
Collecting trend data
The percentage is based on available heat signal, not comment count or independent people.
The evolution of coding agent setups is a topic of discussion, particularly regarding the use of sandboxed environments. Initially, many users ran AI agents directly on their machines. The conversation now explores various setups, including bare metal, containers, VMs, and cloud sandboxes, and the reasons behind transitioning between these environments.
In the early days of coding with AI agents, not that many people were running them inside a sandboxed environment most of us just ran them directly on our machines.
Running coding agents directly on your machine, without any isolation, taught a lot of us a lesson the hard way: it can actually harm your system. Not because the AI is malicious — usually it’s something as a typo in a command that goes sideways.
Personally, I started out running Claude Code/CLI directly on my machine, then moved to a dev container, and now pretty much all of my projects run in the cloud. It’s been good enough so far.
What’s your setup? Bare metal, containers, VMs, cloud sandboxes — and what pushed you to change (if you did)?