The Neo-Reactionary#

Any project who does not lynch any contributor who ever once touched an LLM is slop.

They seem to think resisting AI by just isolating everyone who talks about it1 is going to work. I don’t think these people are going to be very happy going forward–you have developers like Torvalds and Bellard testing out the models and finding the large ones to be good enough at low level work to admit they are useful sometimes. The amount of software that is going to have no interaction with anyone who interacted with Claude is asymptotically zero. That position is a house on a broken foundation that is sinking in to the swamp.

I see people call them Luddites or Neo-Luddites. Though Ludd was more concerned about labor rights and what technology was doing to people. Not really trying to get everyone who ever thought a sewing machine was nice ejected from society.

The Vibe Coders#

I don’t even read the code anymore. I just copy & paste the error back in to the prompt.

These are the people who are either terrible at coding, brain rotted from LLM overuse, or found out how to play the corporate game against their bosses, and just let LLMs write garbage code and cash the cheque.

They’re honestly a bit of a problem in their ability to “write bad code faster.” These are the people that managers in love with artificial idiocy drool over–and is apparently empowering to do even more of it through agent fleets. Now they can spend even more money and resources running prompts through prompts! Somehow this works! Or doesn’t.

You see projects like MemVID2 where their documentation is heinously out of date all over the website and software. One blog says open source. The CLI tells you about “free plans” with a gigabyte limit. The software errors at 100mb. Policies were changed, probably at the agent prompt, and nobody ever went over it all to make sure anything was still true.

Bad devs cope about “moving too fast” to document. Worse devs document everything wrong.

The Utility Minion#

I use Google Gemini at work to write scripts that reboot my routers.

People who basically are unskilled, do not claim to be skilled, and the LLM allows them to create one-off tools that solve some personal need. The code is absolute slop and probably a nightmare. It’s likely nobody understands it. And it’s probably fine that they don’t. This is basically the level of somebody using hypercard3 to make a tool to help run their non-profit4. 10$ in token fees just gave them something they were otherwise completely unable to obtain. I think this is fine. Those people tend to be honest about what they are doing.

I think those people would still be served better by us bringing back the 1990s aesthetics of power user tools, Warcraft 3 trigger editors5, Macromedia Fusion6 (later GDevelop) event sheets, and otherwise visual ways that allow non-programmers to stumble their way in to working systems. An “on ramp” to competency. The LLMs should be helping with “activation energy” or helping you find the right patterns to put here. We’ve ultimately failed this type of people.

The Demings7#

AI is a tool like any other in the box.

These are utilitarians that impose the “100% inspection” rule on LLM generated code. You’re allowed to use whatever bot and method to make the code. You’re held responsible for being able to explain it. You personally are signing off on releasing this code with your name stapled to it. I think this is the more sustainable behavior set.

Looking at the quality of code is ultimately the right metric regardless of how it got made. LLMs can come and go and this behavior set is going to remain completely unphased by it–indeed, some of the rules about refusing large un-discussed changes are just good project hygiene.

The Artificial Idiot Managers#

My message to CEOs right now is that we are the last generation to manage only humans. -Marc Benioff8

These are people who just have some kind of fetish to get rid of all the employees.

They are not simply making them available for use. They are not asking employees to try them and see where they’re a good fit for reducing workplace tedium. This isn’t a constraint solving exercize where human labor is fit somewhere more suitable while robotic labor handles grammar checking. They are absolutely dead set on finally being able to fire that last employee on the team and enjoying the benefits of a fourth industrial revolution.

Problem is industrial revolutions work by providing a product that is actually superior for the operator to use:

  1. A sewing machine does a more consistent job faster

  2. Workshops buy one and clear out their backlogs with it

  3. Other workshops buy them because their competitor has one

  4. Singer makes a lot of money

  5. Industrialists trade stocks and wax their monocles over the relatively choked supply of sewing machines compared to general laborers

The leverage comes as a benny for the benefit. AI isn’t yet shipping the benefit–they just want to rush to the rewards first.

Postword#

We are caught up in strange times of very wealthy people wanting control so badly they are completely exposing their insanity to achieve it. Countries like Japan9 are repealing privacy laws to appease AI scrapers. People are forming batshit anti-AI cults and companies are tripping over themselves to deploy LLMs regardless of fitness. A handful of pragmatic people are getting their use out of the things and getting out.

  1. Seriously, there’s been attempts to force fedi instances to blacklist people just for talking about AI with the unapproved opinion. ↩︎

  2. I had evaluated this one as an interesting test case for small local search engines. Ultimately we did not use it and they did not get their enterprise fees. SQLite and Postgres have suitable extensions to make strong search engines these days. ↩︎

  3. Apple accidentally shipped the original “Rapid Application Development” platform. Hypercard was a mix of powerpoint, SQLite, and the scripting language was the inspiration for Javascript. It inspired Shockwave and the lighter cousin Flash which powered web games for millenial’s childhoods in the 1990’s-2010s. The hit puzzle game Myst was originally a Hypercard deck. Hypercard was so successful at enabling power users to make their own smol database applications it nearly created a rent-seeking crisis. ↩︎

  4. I interned very briefly at a non-profit that built their entire management system around an in-memory database platform for OS X. ↩︎

  5. Valve’s DOTA 2 is based on the DOTA series of maps created for the Blizzard title Warcraft 3. League of Legends is based on DOTA and “League” maps which were also created for Warcraft 3. ↩︎

  6. Baba Is You was made originally in MMF and later ported for console support. Multimedia Fusion launched many an indie game designer’s career. ↩︎

  7. W. Edwards Deming authored Out of The Crisis. His management thesis resulted in post-WW2 Japan’s meteoric economic rise which ultimately informed the cyberpunk trope that Japan was going to buy out the world. ↩︎

  8. "‘White-collar recession’: Benioff says Salesforce won’t hire engineers this year due to AI." The San Francisco Standard. February 27th, 2025 ↩︎

  9. “Japan relaxes privacy laws to make AI development easy.” The Register. April 8th, 2026. ↩︎