It has been a while since I updated folks on what’s new. The newest news is that I have decided to accelerate my augmented development by using Claude’s auto-mode. It makes things a bit less tiresome and it allows me to multitask more in line with how I actually think.
The second part is that I have decided to use Ponytail in order to save my token budget and extend my own conception of what plugins are available and how I might use them. Interestingly enough, I can’t stand to let Claude code do my git work. Then again, I’m not using worktrees or collaborating, so I realize a good part of DevOps is not operating under stress. Everything is in develop and nobody is doing PRs. I don’t even use the term PR.
Item Three: Next-Gen PMO
I have reorganized my major workflows into four buckets.
Metro - Data Engineering Tools & Methodology
Southwall - Local First Development Homelab
Jobber - AI job matching, tracking of job descriptions
Logos - Next Gen Social Media
The idea here is to offload some of the work I’m doing to keep track of everything from my old GTD system of Things for Mac into the more workmanlike Linear for project management. This is going to cost me 12 bucks a month in Linear, but that’s worth it. I’m beginning to write custom dashboards for keeping Linear in sync with Cowork (aka Claudette). This is going to be my mainstream discipline for actual work productivity going forward. It’s not a ‘second brain’ but it will leverage Obsidian as a document base. At some point in the future, I’ll be using Ducklake and/or Postgres for my structuring of unstructured data.
If you’re not familiar with Logos, here’s the high-level and here are some details and here is the video. I’m doing it all myself now. So I need to manage my time and focus in a new way.
Item Four: MDU Update
The working marketing title for the code for md-util is Metapond. I’m putting some more time into it now, following closely on the work of Kurt Cagle with a local-first, code-first, CLI based toolset that will allow me to formalize ontologies for every major data format. I know it’s boring but I am making progress. What’s new is that I essentially have nailed down the cycle, and I’m about to start the process of forward and backward validation on some fraction of my many datasets.
MDU Cycle
init
ingest
catalog
validate
transform
emit
The tough part will be the transform part because I want fully self-documenting business rules. That means I should be able to read something like Essbase MDX transform it into SHACL / SPARQL and know I can validate it with standard and edge cases. This will include IPUMS and Palantir ontologies. My aim is to make some deterministic handles on things that by Shannon and Turing, LLMs should handle very well in ways only a few people truly understand.
In addition, Cagle just started a W3C group and I’m going to eavesdrop quietly and just listen for what they say holons are going to be. Maybe I’ll use it, maybe I won’t.
Item Five: BoxTag MVP
Well, it still has a long way to go, and although my favorite angel investor says it’s a waste of time, I still built a prototype. It looks like this and it works. You’ll remember that BoxTag is multidimensional comment karma. As soon as I saw it work, I realized that I have to anonymize it the first time or else it kind of biases subsequent votes. That’s just a bit of movie magic which I expect a web expert will show me. Note that I used {Authority, Agreement, Emphasis, Relevance}. You can figure out why they are color-coded that way.
Plus One: HomePage
I run so many local webs servers on localhost that it gets confusing to know how many Caddys or Go servers I’m running. So I built this.
