@systemdesignone using a component aware annotation tool when working on UI that will send a screenshot, component source files and annotation straight to an agent with a skill π―
Building a harness that runs in the cloud and can bridge machines by calling SetCWD(<machine_id>, <path>)
This way orchestrating and managing vms becomes much easier without needing a local harness running. Then you can run local agents on the vms like cc/codex. thoughts?
we (among many other custom linting rules) enforce a line limit of 1K lines per file on code. What linting rules have you found most helpful with agents?
@dedene I think it's easier when the product is getting started. AND we had to build custom tooling to get this to work well lol. I guess it's a startup now https://t.co/w1d6LSJpJW
@iannuttall We just setup our own self hosted runners with VPS. codex/cc is great at setting this up in a few minutes and can scale this quite easily if needed. We got away cheaper with a combined 64 cores, 24/7 < 100$ per month
Huge props to @theo . The first video he posted about t3code got us thinking about different ways of interacting with agents and made us build https://t.co/w1d6LSJpJW
@mattpocockuk Been running everything on VMs or remote macs in the office. It's been so refreshing and freeing. No longer constrained by the limitations of local.