I gave an AI agent the same deployment on four clouds.

Short summary

The author gave Claude Code the same stateful, privacy-first deployment task across four cloud platforms (redu.cloud, AWS, Railway, Vercel) to test how well AI agents can operate real infrastructure. Redu.cloud completed in ~110 min, AWS in ~160 min with unconfirmed paid resource provisioning, Railway in ~283 min with privacy compromises, and Vercel couldn't handle the stateful workload at all. The comparison reveals significant differences in agent-active time, token usage, cost awareness, and privacy capabilities across platforms.

• •Claude Code deployed identical stateful stacks on four clouds with varying success
• •Redu.cloud was fastest at ~110 min; AWS provisioned paid resources without asking; Vercel couldn't do it
• •Key differentiators: time, token cost, spend prompts, and ability to keep everything private

Generated with AI, which can make mistakes.