David A. Windham thumbnail

Artificial Intelligence ( Part 5 ) 

I just read my previous four essays on artificial intelligence and this is absolutely the wildest series yet given where it’s started and where it’s going. I laid out my ire towards the word intelligence and the idea in round one1. In round two I joked at how absurd the hype around it was and dipped my toes in the water2. In round three I had already taken the deep dive and started referencing the potential3. By round four, I’d taken to the deep end and started using it daily4. And here we are. 

I use AI daily and not just when I’m working. I photograph plants and birds to identify. I ask about recipes and whatever else comes up IRL discussions. And let me first confess that although I tinkered for a long time with various providers and local models, I now pay a car payment equivalent to Anthropic monthly because I’ve got the API plugged in various places and I’ve got a Max account running in my terminal and editor. I haven’t yet plugged it entirely into my computer ecosystem, but I know it’s just a matter of time. 

The first essay I published on AI was in July of 2022 and I had likely been tinkering with it for the year prior. Last year I started involving AI at the beginning of my projects and this year, I’ve plugged it into the entire workflow of two projects. I say entire, because I’ve started building what I think it’s a superb workflow. What really turned it around was building custom Model Context Protocol servers earlier this year because they allow me to control access and abilities on a very granular scale5.

I’m now able to give the agents enough context to essentially act as a pair programming partner and coworker. For every set of tasks within a project, I’ll define a set of parameters using a single directory of markdown files within the project. I keep similar files stacked together with a decent naming convention and I ask the agent to keep up with them. I’ll keep a preview of the markdown files in a window so that I can review changes. I’m not down with the agentic Ralph style of automation yet, but I don’t require the agents to ask for edits anymore. I save every script written with a naming convention similar to the tasks list. The overview markdown has internal links to each script and sub-project files acting essentially like an Obsidian vault for the ‘second brain’ folks out there6. I’m sure I’ll be tweaking this workflow as I move more projects through it but it’s all built around the same thing I’ve been doing for years. 

For what seems like eons, I have been using automated tools. I use and write scripts to automate processes. Before 2000 I had figured out how to use Visual Basic in Excel and after I was scripting photoshop exports for projects. By the mid 2000s, I was using lint checkers to verify code and building software as combinations of scripts to automate things. I say ‘things’, because that’s essentially what it all is – the same software that allows users to log in and log sales to a spreadsheet is no different than the pre-2000 Visual Basic. All of these are tools for automation and AI is absolutely no different. Instead of me running a project building features and stopping to run the linting tool as an editor plugin, I’m just automating that into the workflow with the AI workflow list. I remember when CI ( continuous integration ) was the big thing and this is essentially the same, but instead of running a bunch of Jenkins scripts and configuration files, I’m using the same python and bash scripts I’ve used for twenty years. 

I follow a bunch of developers on the socials and I love to watch them knock on AI for various reasons. I do believe that the ‘hype’ and money pouring into is a little like every other ‘whizamajig’ that’s come before. My favorite meme on the subject is this one.

couldn’t find a source on this or I’d cite it

Now my first and last interactions on ‘the machines’ ( that’s what I call the computers ) are almost always via a chat log with an Ai agent. I’ll ask what’s on the task list while still drinking coffee outside watching the dog run and I’ll ask to summarize the days’ work right before I hit the sleep button. I’m not idiot enough to ask psychosis inducing questions about how to live my life or assign any consciousness to the agents. But I’ve long been anthropomorphizing the machines7 so we shall see what happens when I start plugging it into my communications, bank accounts, grocery, and todo lists. I’m sure part 6 is forthcoming. 

  1. Artificial Intelligence – https://davidawindham.com/artificial-intelligence/
  2. Artificial Intelligence ( Part 2 ) – https://davidawindham.com/artificial-intelligence-2/
  3. Artificial Intelligence ( Part 3 ) – https://davidawindham.com/artificial-intelligence-part-3/
  4. Artificial Intelligence ( Part 4 ) – https://davidawindham.com/artificial-intelligence-part-4/
  5. Model Context Protocol – https://davidawindham.com/til/posts/mcp
  6. A Second Brain – https://davidawindham.com/a-second-brain/
  7. Anthropomorphizing Machines – https://davidawindham.com/anthropomorphizing-machines/