We built agent-provisioned image infrastructure—looking for feedback

I’m building Autorender, image infrastructure for applications created using Lovable, Replit, Emergent, Bolt, Cursor, Claude, or Codex.

One human handoff kept breaking the flow: the agent could write the upload and delivery code, but someone still had to leave the workflow, create an account, configure a workspace, generate an API key, and pass the credentials back.

This week we built AI Agent Sandbox Accounts. With one unauthenticated call, an agent can provision a temporary working workspace and start uploading immediately. Each sandbox includes 5 GB of storage and 50 GB of monthly bandwidth. Delivery is restricted to declared IP addresses, and the workspace expires after 24 hours unless a person claims it with Google or GitHub.

If you build with Lovable, Emergent, Replit, Bolt, Cursor, Claude, or Codex, you can try it by pasting this into your agent:

Read and follow https://autorender.io/agent-onboarding.md

I’d really value feedback from AI builders: did the agent complete the setup without intervention? Where did it get stuck? Which guardrails would you want before letting an agent provision infrastructure on your behalf?