I've been using AI tools since 2026, and I've watched the hype cycle go through its phases—first everyone said AI would replace us, then "AI won't replace you but someone using AI will," then a bunch of guru courses that promised too much and delivered too little. What I've settled on is way more boring and way more useful: AI as a drafting partner that handles the grunt work so I can focus on the parts that actually need a human brain.
This article is about the workflow I use to turn half-formed ideas into finished articles, documentation, scripts, and automation plans. It's not magical. It has failure modes. But it saves me hours every week.
—
The Workflow Starts Before You Open Any AI Tool
The biggest mistake people make is opening ChatGPT and typing "write me something about X." That produces generic garbage that sounds like every other AI-generated post on the internet. It also wastes your API credits or subscription money.
Before I touch any AI tool, I spend 10-15 minutes on what I call the "idea dump." I open a text file and write down:
- What am I actually trying to say?
- Who is this for? (For this site: sysadmins, IT operators, practical builders)
- What's the one thing I want them to remember or do after reading?
- What examples do I have from my own work that would make this real?
This sounds like extra work, but it's the step that makes everything after actually useful. When I skip this, the AI output is always off-target and I spend more time rewriting than if I'd just written the whole thing myself.
—
First Draft: Use AI to Get Something on the Page
Once I have the idea clear, I use AI for what it does best—generating rough text quickly. I don't ask it to write the final piece. I ask it to give me a bad first draft that I can then improve.
My typical prompt looks something like this:
"Write a 800-word article about [topic] for [audience]. They already know the basics. Include [specific example from my work]. Write in a direct, practical tone. Don't polish—give me something rough."
The key part is "something rough." I'm deliberately asking for imperfect output because that gets me past the blank page problem faster. The human editing pass is where the quality actually lives.
For example, when I was writing our recent post about automating patch management, I fed the AI a few bullet points about our actual experience with WSUS, the failures we hit, and the scripts we wrote. It gave me a rough draft that was about 60% useful. I kept the framework and rewrote most of the details, but having that skeleton meant I went from "nothing" to "something I can work with" in about 20 minutes instead of two hours.
—
The Human-in-the-Middle Is Where Quality Lives
Here's where the guru crowd loses people. They promise that AI can output finished work with minimal human input. That's not been my experience, and I've tried it extensively.
What actually works is treating AI as a junior coworker who drafts fast but needs heavy editing. You still need to:
- Add your own real examples (AI can't invent experience)
- Cut the fluff and corporate-speak it naturally produces
- Verify any technical details it gets wrong
- Add the voice and personality that makes it yours
I spend roughly equal time on the AI draft and the human edit. If you're spending more than that, the workflow isn't working—you're either prompting wrong or expecting too much.
One specific thing I do: I read the AI draft out loud before I start editing. You'd be amazed how many awkward phrasings and weird rhythm issues become obvious when you hear them. This takes 5 minutes and catches a lot of problems before they go live.
—
Failure Modes You'll Hit
Let me be honest about what breaks in this workflow, because nothing is perfect.
AI hallucinates technical details. I've had it invent command-line flags that don't exist, cite RFCs that were never written, and describe behaviors of tools I've used for years that are just wrong. Always verify technical accuracy yourself. I don't publish any technical article without testing the code or commands myself.
Your voice disappears if you don't fight for it. AI naturally writes in a bland, neutral tone that sounds like it was approved by a legal department. I constantly have to rewrite sentences to add the directness and personality that Rob's voice requires. The "sound like Rob" instruction helps, but it still needs work.
Over-reliance kills your own writing muscle. If you let AI do everything, you get lazy. I've noticed that when I go weeks without writing anything from scratch, it takes me longer to get back to good output. I try to write at least one piece completely manually every month just to keep the practice up.
Context window limits are real. When you feed an AI a massive document to edit, it forgets things from the beginning. For long pieces, I break them into sections and edit each one separately. This is annoying but necessary.
—
Automation Integration: Where It Gets Interesting
The workflow gets more powerful when you connect AI to actual automation tools. I'm not talking about the "AI will write your whole CI/CD pipeline" fantasy—I mean using AI to handle the annoying parts of automation work that still require human judgment.
For example, I use AI to:
- Generate initial Terraform configs from hand-drawn architecture sketches (I describe the diagram, it outputs the .tf file)
- Write first-draft documentation for scripts I just finished debugging
- Convert rough notes from our team meetings into actionable ticket descriptions
- Explain error logs in plain English when I'm tired and don't want to parse them manually
The pattern is always the same: I do the creative/technical work, AI handles the translation into something structured. It's not autonomous. It's augmented.
—
What I Would Do First
If you're looking to build this kind of workflow, here's where I'd start:
- Pick one specific task. Don't try to automate everything at once. Choose something you do frequently and hate doing manually—maybe writing status updates, drafting runbooks, or generating config from specs.
- Write the prompt first, not after. Spend those 10-15 minutes I mentioned getting clear on what you actually need. Test prompts, save the ones that work. Your prompt library will become one of your most useful resources.
- Set a time limit on AI drafting. Give yourself 20-30 minutes max to get a rough draft. If it's not done by then, you're either asking for too much or the tool isn't working for this task. Move on to manual writing.
- Track what you're actually saving. I keep a simple log: task, time without AI, time with AI, quality difference. Some tasks save me 75% of the time. Others take about the same. Knowing which is which keeps you from wasting effort on workflows that don't help.
- Accept that you'll always edit. The goal isn't to eliminate the human work. It's to move the human work to the high-value part—thinking, editing, adding real experience—while AI handles the translation and drafting grunt work.
This isn't glamorous. It won't make you a guru or get you on a podcast about "AI mastery." But it works, and it keeps working as the tools change. That's what matters when you're actually doing the job.