My Journey of Building a Terraform AI Agent — Automating Cloud Infrastructure with AI

31 October, 2025

By Anoop Kumar, Lead DevOps Consultant, Rackspace Technology

The vision: Why I built this AI agent

As a DevOps Engineer who is passionate about AI, I always seek ways to optimize and automate cloud infrastructure. Managing Terraform code manually is time-consuming, error-prone and requires constant validation. I wanted to build something that could help my team and customers deploy resources faster, smarter and with fewer mistakes.

I shared my idea with my manager, Marc Bloom, who encouraged me to explore the idea further. I also sought feedback from my colleagues, who helped me refine my approach. This led to the creation of the Terraform AI Agent, a generative AI-powered tool that generates, validates and deploys Terraform code in Microsoft® Azure® automatically.

The journey: From idea to execution

Two months ago, I started researching how to automate Terraform with AI. Initially, my goal was simple: Generate Terraform code automatically and save it into .tf files.

The challenge: AI hallucination and duplicate code

However, I soon faced a major issue. AI was generating the same code repeatedly for similar requests, leading to duplicate resources in Azure. This could create conflicts and unnecessary costs.

The solution: Smart code generation and validation

To fix this, I implemented a function that checks existing Terraform infrastructure before generating new code. This helps ensure:

No duplicate resources are created
Unique and necessary code is generated
Custom modules and templates can be used to meet compliance requirements
AI hallucinations (incorrect code generation) are eliminated

After solving the code generation problem, I focused on validating and deploying the Terraform code seamlessly

How the Terraform AI agent works

Step 1: Enter infrastructure request

Instead of manually writing Terraform code, users enter their infrastructure request in a simple prompt, including:

Example: “Create a resource group named aiagent-terraform-rg in uksouth”
The AI instantly generates Terraform code while checking for existing infrastructure to prevent duplicates
Supports custom modules and templates to ensure compliance

Step 2: Validate and auto-fix code

Before deployment, the AI performs multiple checks:

Runs terraform fmt to format the code
Uses terraform validate to ensure correctness
Applies TFLint to detect security misconfigurations
Auto-fixes errors to maintain best practices

Step 3: Deploy via GitHub Actions

With a single click on “Save & Deploy,” the Terraform AI agent pushes validated Terraform code to GitHub and triggers GitHub Actions, which run:

terraform init
terraform validate
terraform plan
terraform apply

It then deploys infrastructure in Azure and stores the Terraform state in Blob Storage.

Terraform AI agent architecture and workflow

GitHub repository: https://github.com/anoopkum/terraform-automation

Step-by-step implementation

The heart of our solution is the AI agent that converts natural language into Terraform code. Looking at the repository, we can see how this is implemented in aiagent.py.

Initially, I use OpenAI’s GPT-4o model. But I later switched to O3-mini, which is good at reasoning/code to convert natural language prompts into structured Terraform code.

User interface workflow

UI has three main action buttons that guide users through the infrastructure creation process:

1. Generate Terraform code button

When users enter their infrastructure requirements as a natural-language prompt, they can click the “Generate Terraform Code” button. This triggers the AI agent to:

Process the natural language description
Convert it into properly structured Terraform code
Display the generated code for review

The generated code appears in a code editor interface where users can review it and make any desired adjustments.

2. Validate Terraform code button

After generating the code, users can click the “Validate Terraform Code” button to initiate the validation process, which triggers:

Code formatting with terraform fmt
Syntax checking with terraform validate
Best practices verification with tflint

Green checkmarks indicate successful validations, while red alerts highlight areas needing attention.

3. Save & deploy button:

Once the code passes validation, users can click the “Save & Deploy” button to initiate the deployment process, which:

Saves the code in a terraform file (main.tf)
Creates a new pull request in the GitHub repository
Triggers the GitHub Actions workflow
Automates the approval and merge process
Deploys the infrastructure to Azure

The interface provides real-time status updates during the deployment process, showing the progress from plan generation to resource creation.

Terraform state management

A critical aspect of Terraform infrastructure management is the state file. In our architecture, the Terraform state is stored securely in Azure Blob Storage.

The GitHub Actions workflow initializes Terraform with backend configuration that points to the Azure storage account and container. The storage account access key is securely fetched from Azure.