Created: 2025-06-28
Last Updated: 2026-04-05
Stage: budding
AI
Where I’m At
I’m still highly skeptical of generative AI, but I am seeing useful application of the technology. I’m primarily focused on using AI from the perspective of a programmer and engineering management. Your mileage may vary. As of November 2025, the frontier models have become pretty good at generating code. Given a reasonably detailed specification, the current models can create working programs. To date, I have been using AI to build small scripts and to augment my own abilities. I have used it to build a Python script that will diagram the structure of an AWS organization, including the organizational units and the subaccounts. I also use it to create jq expressions instead of struggling to build them myself. I find it especially helpful for things I only use occassionally, and for which I burn lots of time relearning how they work each time. I’v even used it to analyze spreadsheets to help me clean them up. (Though I wasn’t confident enough to let the AI perform the final edits) I continue to be frustrated by the AI hype machine. I don’t believe AI will be taking our jobs any time soon. While the AI can propose and implement solutions, there is still a need for a human to review and ensure the AI’s work makes sense within context. I am also frustrated with the AI naysayers as it clearly has useful applications. Between the two extremes lies a productive middle.
Since I have had good results so far, I want to experiment with more advanced techniques. I want to try removing more of myself from the loop and let the agents perform larger tasks. As a part of that experiment, I want ot try working with subagents, taking a large task and having the AI break it up into smaller pieces and hand off each piece to a an agent running unde the main controlling agent. I am also interested in testing out some of the current generation self hostable models.
I am primarily focused on using AI for work purposes. I experiment with it for personal projects too, but I am reluctant to become dependent on it. Due to the vast amounts of data and compute needed to build and train a model, I don’t see a world where it’s feasible to build a truly open model. Currenlty the cost to use a frontier model is someewhat reasonable due to the costs being subsidized by investor capital. At some point that party is going to end and I expect prices will sharply increase.
Concerns
- Only mega corporations with deep pockets or well-heeled investors will be able to build these models
- What happens when AI tools stop being subsidized by investor money and start charging market rate?
-
Where will the next generation of programmers come from if everyone is using AI to write code?
- Once the current generation of senior+ developers ages out, where will we find people who can understand and debug the programs the AI writes?
-
How will the practice of programming be moved forward?
- AIs don’t “think” or “reason”, so I don’t see how they can innovate something completely new
Things I Find Interesting
-
CLI interfaces for AI platforms
- Gemini
- Claude Code
- Subagents
- Agent orchestration
- Self hosted models
