My Successes with AI-augmented Code as of March 2026

- 4 minutes read - 835 words

Being a computer dork of a certain stripe, I’ve had a TODO(.md) in my personal wiki for a long time that was full of IT / tooling tasks that I wanted to do, but that I never found the time:

  • Port the NeXT desktop to FreeBSD
  • Clean up my blog repository of incorrectly committed binary assets that were bloating out my storage footprint and slowing up checkouts
  • Remedy data loss in my blog due to a botched cleanup operation
  • etc.

I delivered every one of these with the help of AI. So I thought I’d take a quick inventory from roughly last year at this time to now.

The List

  1. Ported a Linux-only desktop environment (90% working ;) ) to FreeBSD. Technically I could have done it without, but good prompting and a decent tech background let me get there: “NextSpace on FreeBSD”
  2. Built an ingestion pipeline of recipes from printouts, photos, and spam sites into local, ad-free Hugo blog posts
  3. Rewrote my blog git repo to remove bin assets (20+ years of content!). I had been using Hugo’s page bundle feature as I migrated and normalized formatting, etc. from Movable Type, Wordpress, and Octopress across this blog’s history
    • Built the NGINX configuration to serve them
    • Deployed the “Let’s Encrypt” SSL certificates
    • Built hugo.toml configs in an efficient “mergeable set” so that I have the ability to blog on laptop, on server, on Mac
    • Reduced blog’s git footprint by two orders of magnitude: 700MB to 7MB. Also, (image)magick-ed all the assets to web-sensible sizes (I’ve never known how to calculate that for various formats, etc.)
    • Built a “create new post” skill that makes adding images (in the asset dir) and doing the creation / templating / etc. of the new post a snap. It’s as easy as it was in WordPress but with the vim-based hackeritude
    • (In progress) A voice-based blogging / newsletter writing interface that will allow me to return to the joy of writing and scalably communicate with family again by turning my walking commute time / phone idling time into something more meaningful
  4. I botched my “Clean up WordPress imported posts into Hugo” IT chore 5 years ago. About six years of posts had had their final paragraph chopped off thus losing the punch. Claude walked the affected posts, showed me the differences, and helped me get my content back thus saving me hours of tedium and / or coding / debugging. Sure, I could have built a while loop and done some Python to find the posts affected and present vimdiff…but it would have been buggy and, uh, I messed it up last time. Claude Code one-shotted this. Fixed in a toddler’s nap duration.
  5. Normalized blog links from Jekyll, Octopress, Hugo, Wordpress thus fixing broken links across 15 years of content (hundreds of posts)
  6. Built a cross-platform (!) phonics app for my son who’s learning to read (tested on FreeBSD 15.0-CURRENT, Linux Mint)
  7. Evaluated a number of laptops for FreeBSD fitness. On FreeBSD, hardware support is not (yet) guaranteed and Claude helped me assess whether the problem was me and the OS or lack of support. Short aside: buy a Thinkpad.
  8. Blog post copyediting and readability feedback
  9. Porting my vim config to neovim; establishing usability parity with a smaller plugin set; getting beautiful colors for feedback
  10. Configured a Common Lisp programming environment via neovim + conjure + swank
  11. Etc.

The Context

So these are places where I, personally have seen value added to my life by these tools. I don’t understand how others aren’t getting this value; I don’t understand why so many people, even experts, maintain “this can’t do anything good.” Are they not doing what I did? I’m more sympathetic to avoiders who do so on an ethical stance. Let me start with them.

There exist those for whom AI is a no-go. They claim that it’s a tool of fascistic overlords; they note the large-scale copyright theft implicit in the models’ training sets; they note the insane environmental impact; etc.

I think we’ll figure out the energy situation (perhaps leading to a revolution in energy policy). When the AI lobby can tell the fossil fuel lobby “Shush, we need this geothermal / wind / solar / fusion reactor / Dyson sphere,” we may be in a vastly better world.

On the copyright front, I’ll wager that it’s more likely that industries will seek remedy in court rather than anyone turning back this tide like Canute with a sword at the shore.

There are also those who claim that AI can’t generate new answers and/or can’t do anything useful. It might well be the case. Maybe buckets of tokens won’t generate a better C compiler or fold a protein correctly so as to discover a new drug. I can grant this.

But I also noted that many of us — IT professionals especially — need something more capable than Automator, Apple’s visual work pipeline constructor kit that held oh-so-much promise, and I hope my list above attests to that.