I use AI in almost everything on this site and in my GitHub repos. This page says where, and what stays my job.
The short version: I ideate and build with my AI tools. They know a lot about me: how I work, how I learn best, and where I'm weak. AI helps me move faster, but I'm using my own brain at every step. I do the work. They assist.
Building
Most of my repos have major parts I wrote by hand. Cursor and Claude worked as tutors. They explained the pattern and described what the code needed to do, I typed it, and they reviewed what I wrote.
The builds ran on a two-agent workflow. One agent wrote code from a spec I approved, and a separate agent audited the diff before anything was committed. No agent graded its own work.
Once I understood what I was looking at, whatever the language or tech, and trusted the tools to build within my constraints, I gave Cursor and Claude more autonomy. Cursor builds from a spec I write and approve first, so the scope is mine before any code exists. Claude Code reviews each pull request on its own, separate from the agent that built it. I read both, then fix, reject, or merge.
The same setup covers my cloud work. My AWS labs and Terraform run through the same spec, build, and review loop as the repos.
My merge rule is simple. If I can't explain a piece of code in one sentence, it doesn't ship. Simpler code I can defend beats clever code I can't.
Writing
My LinkedIn posts start as me talking. I dictate with Handy or I give Claude a short brief. Claude drafts in my voice, and I edit before anything goes out.
The ideas and the experience are mine. Claude organizes and tightens. If a draft claims something I haven't done, the line gets cut.
This page went through the same process.
Research
I use Claude to find and summarize sources. When a claim matters, like what a control requires or what a framework actually says, I check it against the primary source: the NIST catalog, FedRAMP and CJIS documentation, or the vendor's own docs.
For video courses and talks, HoverNotes captures my notes while I watch. Claude turns the raw capture into a study note, and I keep what's worth keeping.
Automation
I run a small set of Grok Bots for work that repeats. They help with career planning, brand building and social media management, personal training, and turning long notes into short summaries.
Each Bot has a written job description and a list of things it may not touch. Anything that would post or send in my name asks me first, every time, on every tool.
Their output is labeled as theirs. A Bot's summary never gets passed off as my writing.
I'm also testing other personal assistant agents, Instinct and Muse right now. Neither acts in my name yet.
Principles
- I review everything before it goes out, and I own it after.
- AI involvement is shown, not hidden.
- Nothing is AI-free, but everything is still mine.
- Plain language. If I can't say it simply, I don't understand it well enough to publish it.
Last updated 2026-09-27.