5a734d404b
- Add self-hosted GoatCounter via systemd binary service (stats.monotrope.au) - Add Miniflux RSS reader via Docker Compose (reader.monotrope.au) - Extend Ansible playbook with goatcounter and miniflux tags; all provisioning is idempotent - Add Caddy reverse proxy blocks for both new services - Inject GoatCounter script in baseof.html (production builds only) - Add goatcounter and miniflux Makefile targets - Rewrite CLAUDE.md to reflect actual project state Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2.1 KiB
title, date, draft
| title | date | draft |
|---|---|---|
| Thoughts on AI, circa April 2026 | 2026-04-08 | true |
Part of the reason I wanted to make this site is to practice writing more, and in part that's because the writing I do at work (project plans, technical outlines, etc.) are increasingly written with the help of AI.
This opens up a can of worms, of which some I will attempt to dissect here. A disclaimer: I don't think my views on this topic are currently ideologically consistent. Broadly I think this is ok, and I think with nascent and rapidly changing areas it's valuable to be able to hold opposing views in your mind for awhile. As the title of the post indicates, these thoughts are also current as of mid-April 2026, and are very likely to change dramatically.
Firstly, I think AI is, in many contexts, useful. I have spent a bit of time on Bluesky and a lot of anti-AI sentiment there revolves around "AI is dumb, it makes stuff up, no one actually gets value from it." As someone who works in software it is incredibly obvious to me that AI for coding is increidbly useful if you know what you're doing. But I think AI can also be genuinely useful for writing, researching, and ideating. Some of the "AI is not useful" critique will be rooted in a view of AI that is a year or more out of date. "Asking ChatGPT for something instead of Googling for a simple answer" might have been dumb when dealing with early versions of the models that couldn't use tools or search the web, but an AI that can run multiple web searches and consolidate the results is a real value-add, especially when you're not wading through pages of ads and fluff in the search results.
However, AI in the workplace has not been all sunshine and rainbows. The teams are writing code more quickly, routine work like version upgrades is much quicker, and my personal Claude/Obsidian setup has been good for my own context and productivity. But holy shit the documents. You get out of a meeting where someone takes an action to write a plan and BAM! 7 minutes later they've published a detailed, polished plan that you have to read.
But the issue