Are you letting agents run infra tools / scripts yet?
Posted by shashahchk 10 hours ago
Curious how people are using AI agents in their devops/infra workflows.
For example: • running migrations • deploying services • provisioning infra
Or are you mostly keeping AI in the “code generation” lane?
Also curious if anyone is using MCP servers from tools like Supabase or Vercel.
Would love to hear how you are shipping today.
Comments
Comment by pmf_hunter 1 hour ago
For infra specifically: agents can run shell commands, manage Docker, deploy via git. The key insight was giving them narrow permissions per role rather than full root access. The research agent can read logs but can't deploy. The content agent can write files but can't touch infra.
Biggest risk isn't agents breaking things - it's agents confidently doing the wrong thing without telling you. So every destructive action requires confirmation, and all changes go through trash instead of rm.
Haven't used MCP servers much - direct tool access through the orchestration layer turned out to be simpler and more reliable.
Comment by fallonshoulders 4 hours ago
Comment by shashahchk 1 hour ago
Comment by verdverm 9 hours ago
Just have them help with IaC
Comment by shashahchk 6 hours ago
Comment by verdverm 36 minutes ago
Comment by rvz 10 hours ago
Comment by jamiemallers 6 hours ago