Using AI assistants in everyday IT work — without switching off your brain

AI assistants have quietly become part of my daily toolkit — Claude, GitHub Copilot, ChatGPT, and lately connecting tools together with MCP. They make me faster, but only because I treat them as an assistant, not an oracle. Here’s where they actually earn their place, and the guardrails I keep.

Where they genuinely help

• Making sense of logs and errors. Pasting a cryptic stack trace or an unfamiliar error and asking “what does this likely mean, and what would you check first?” is a great starting point for analysis — it gives me hypotheses to confirm, not conclusions to trust.
• Scripting and one-off automation. Small Python or shell scripts for parsing exports, reshaping data, or batch checks. The assistant drafts, I read every line and run it on safe data first.
• Writing things up. Turning rough notes into a clean knowledge article or a clear hand-off to 3rd level. Documentation is where good tools save the most time.
• Learning faster. When I hit an unfamiliar concept, a good explanation on demand — followed by checking it against real sources — beats bouncing between ten browser tabs.

The guardrails

This is the part that matters, and the part that’s easy to skip.

• Never paste sensitive or production data. Customer data, internal system details, credentials — none of it goes into an external tool. I anonymise or reconstruct a minimal example instead.
• Verify before you trust. Models are confident even when they’re wrong. I treat every answer as a draft to be checked, especially anything touching a production system.
• Understand what you ship. If I can’t explain a script the assistant wrote, it doesn’t run. The goal is to move faster on things I understand, not to deploy things I don’t.
• Keep your own skills sharp. The assistant is a multiplier on what you already know. Lean on it for speed, not as a replacement for actually understanding the system.

Why I bother writing this down

In a field that’s moving this fast, “I use AI tools” isn’t a flex — how you use them is the real signal. Used carelessly they create confident nonsense and leak data. Used with judgement they free up time for the work that actually needs a human: understanding the system, weighing trade-offs, and taking responsibility for what goes live.

That balance — fast with tools, careful by default — is exactly how I try to work.