Quick Answer

Describe your infrastructure in plain English with the cloud provider name, and AI writes Terraform, Pulumi, or OpenTofu code. Always review IAM policies, security groups, and state backend configuration.

AI knows major providers well (AWS, GCP, Azure, Hetzner, DigitalOcean)
Modules, remote state, and workspaces need human architecture decisions
Never run terraform apply from AI output without terraform plan review

What You'll Need

Cloud provider account with API credentials
Terraform or OpenTofu installed (tofu is the FOSS fork)
State backend (S3, GCS, Terraform Cloud, or local for dev)
AI IDE or CLI

Steps

Describe desired state. Prompt: Write Terraform for a Hetzner CAX41 VPS, floating IP, and firewall allowing 22, 80, 443.
Organize with modules. Ask: Refactor into reusable modules with clear input variables.
Add remote state. Prompt: Add a backend block using S3 with DynamoDB locking.
Parameterize environments. One root per env (envs/prod, envs/staging) with shared modules.
Run terraform plan. Review every resource carefully.
Secrets management. Inject via TF_VAR_* env vars or SOPS. Never commit .tfvars with secrets.
Apply with approval. CI pipeline: plan on PR, apply on main merge with manual approval gate.
Drift detection. Schedule nightly terraform plan — alert on drift.

Common Mistakes

Hardcoded secrets in .tf. Use a secrets manager.
count instead of for_each. for_each is safer for refactors.
Overly permissive IAM. AI defaults to * actions. Narrow manually.
No tags. Every resource should have environment, owner, cost-center tags.

Top Tools

Tool	Use
Terraform / OpenTofu	IaC runtime
Infracost	Cost estimation in PRs
tflint	Lint
Atlantis	PR-based workflow
Pulumi	Typed IaC (TS/Python/Go)

Conclusion

AI removes the cold-start problem of IaC. Start from a rough prompt, harden with a human security review, lock with a remote state backend. Misar Dev scaffolds Coolify + Hetzner Terraform setups in minutes.