AI Is My Daily Co-Pilot — Not a Replacement

A year ago I was skeptical. Today, I can't imagine my workflow without AI. Not because it writes perfect code — it doesn't — but because it eliminates the 80% of routine work that used to eat my day. I'm an infrastructure engineer, not a typist, and AI finally lets me focus on architecture and decisions instead of syntax and boilerplate.

Here's exactly how AI fits into my real, daily DevOps practice — with specific tools, use cases, and honest limitations.

1. Terraform & Ansible: From Idea to Working Code in Minutes

When a client needs a new AWS environment, I used to spend 2-3 hours writing Terraform modules from scratch. Now my workflow looks like this:

  1. I describe the architecture to Claude: "VPC with 3 subnets, ALB, 2 EC2 instances in ASG, RDS PostgreSQL 16 Multi-AZ, S3 for backups"
  2. Claude generates a complete Terraform module with security groups, IAM roles, and outputs
  3. I review every line, adjust to the client's naming conventions and tagging policies, and apply

The key is step 3 — I never apply AI-generated infrastructure without a thorough review. AI doesn't know your security policies, cost constraints, or compliance requirements. But it gives me a solid starting point that's 80% correct, and I refine the remaining 20%.

For Ansible, the workflow is identical. "Write a playbook that hardens Ubuntu 22.04: disable root SSH, configure fail2ban, enable ufw with only ports 22, 80, 443, set up unattended-upgrades." The result is a production-ready playbook in seconds instead of 40 minutes.

2. Claude Code: Building This Entire Platform

This website — the one you're reading right now — was built with Claude Code as my pair programmer. The entire e-commerce platform: product catalog, shopping cart, Monobank payment integration, admin panel, blog, portfolio, SEO optimization, Docker deployment. All developed in a fraction of the time it would take solo.

My workflow with Claude Code:

  • Architecture discussions — I describe what I want, Claude suggests the approach, we iterate
  • Feature implementation — Claude writes the code, I review and test
  • Debugging — when something breaks, I share the error and context, Claude pinpoints the issue
  • Refactoring — "This controller is too fat, extract a service object" → done in one pass

The result: a full Rails 7 application with 48 services, 8 categories, payment processing, 3-language i18n, admin panel — deployed to AWS with CI/CD. What would normally take weeks took days.

3. Incident Response: Faster Root Cause Analysis

When a 3 AM alert fires, the last thing you want is to spend 30 minutes reading through logs trying to find the needle in the haystack. Here's my actual incident workflow:

  1. Alert fires in Telegram from Grafana
  2. I SSH into the server and grab the relevant logs
  3. Paste the logs into Claude with context: "This is an Nginx + Rails app behind ALB. Started getting 502 errors at 03:12. Here are the last 50 lines of nginx error log and rails production log"
  4. Claude identifies the pattern: memory leak causing OOM kills, Rails worker restarting, connection pool exhaustion — whatever it is
  5. I fix the root cause, not just the symptom

This has cut my mean time to resolution (MTTR) by roughly 40%. The AI doesn't replace my judgment, but it's incredibly fast at pattern-matching across hundreds of log lines.

4. Documentation & Runbooks: The Boring Work, Done

Let's be honest — nobody enjoys writing documentation. But every well-run infrastructure needs runbooks, architecture diagrams descriptions, and onboarding guides. AI makes this painless:

  • Runbooks — I describe the procedure ("how to restore PostgreSQL from S3 backup"), AI generates a step-by-step runbook with verification steps
  • Post-mortem reports — I feed it the timeline and facts, it structures a proper blameless post-mortem
  • Client reports — after an infrastructure audit, AI helps me turn raw notes into a professional report with recommendations

The time I save on documentation goes directly into actual engineering work.

5. Custom AI Automations for Clients

Beyond my own workflow, I'm building AI-powered automations for clients:

  • RAG systems — AI that answers questions from a company's internal knowledge base (Confluence, wikis, runbooks). Engineers get instant answers instead of searching through 500 pages of docs
  • MCP integrations — connecting Claude to client's Jira, GitHub, and monitoring tools so it can pull context automatically during troubleshooting
  • Automated triage — AI that analyzes incoming alerts, correlates them with recent deployments, and suggests probable cause before a human even looks at it

This is where AI in DevOps is heading — not replacing engineers, but giving them superpowers.

What AI Can't Do (Yet)

After a year of heavy daily use, here are the hard limits I've found:

  • It can't access your systems — AI works with the information you give it. It can't SSH into your server, check the actual state, or verify its suggestions work
  • It hallucinates confidently — especially with specific CLI flags, API parameters, or version-specific features. Always verify against official docs
  • Security decisions require human judgment — AI can suggest a firewall configuration, but only you know your threat model, compliance requirements, and risk tolerance
  • Architecture is still a human job — choosing between monolith and microservices, selecting the right database, designing for scale — these decisions need experience and business context that AI doesn't have

My AI Stack in 2026

ToolWhat I Use It ForDaily Use Claude (Opus)Complex reasoning, code review, architecture discussions, long documents3-4 hours Claude CodePair programming, feature development, refactoring, debugging2-3 hours GitHub CopilotInline code completion, repetitive patterns, boilerplateAlways on ChatGPTQuick questions, brainstorming, second opinion30 min Custom MCP serversConnecting AI to client tools (Jira, GitHub, Slack, monitoring)Per project

Getting Started with AI in Your DevOps Workflow

If you haven't started using AI in your infrastructure work, here's my advice:

  1. Start with documentation — it's low-risk and the time savings are immediate
  2. Move to code generation — Ansible playbooks, Terraform modules, shell scripts. Always review before applying
  3. Use it for debugging — paste error logs and describe your setup. You'll be amazed at how quickly it spots issues
  4. Never blindly trust the output — AI is your assistant, not your boss. Every suggestion needs human verification

AI has made me a significantly better and faster engineer. Not because it does my job — but because it handles the tedious parts so I can focus on what actually matters: making reliable, secure, cost-efficient infrastructure.

Want to integrate AI into your DevOps workflow or build custom AI automations for your team? See my work or book a free consultation.