How do you solve security when letting an AI deploy full-stack code?

Hey Product Hunt! I'm Tahir, building a self-hosted AI IDE called FikirCode. We are launching in less than 8 hours! While designing our agentic loop, our biggest challenge was backend isolation. We solved it by spinning up restricted, read-only Docker sandboxes for each deployment. How do you usually handle user-generated code execution safety? Would love to hear your thoughts before we go live!