I don’t have an answer to why today’s AI empowerment of non-coders is different from the hope, years ago, that technology should be simple enough that those unfamiliar, those not “geeks”, would be comfortable using it.
Years ago I created an Android app that was in a very small way my attempt at a visual programming language. It allowed a user to create an Android app within their phone with no external assistance or compiler. For the non-programmer, you could see the flow of pages based on button selection, and create branching, looping, and conditional choices without writing code. For the semi-programmer, you could see the C-derived programming language that drove the app, write code manually, step-debug, and inspect live variable states to your (admittedly simplistic) content.
My impulse was: we have in our phones a device that can connect to the web, calculate as much as a desktop computer, connect to a cell network and GPS and navigation, provide visual and audio access. And yet people are left to have others sell access to an immensely powerful sensory/logical/environmental device to them in order to access those features. A phone is a universal tool there for you to use. Use it as you want.
I would ask how is this different than the current vibe coding craze, but it’s a bit obvious even ignoring my cumbersome UI (though, many smarter people have done this many times more clever-er so me presenting my example is merely… vanity). It’s obvious to say, but there’s a compelling value to providing people the ability to use the depth and versatility of the internet to create a tool to allow them to enjoy the simplification of a mundane, difficult, niche task. “I have a garden in [this city]. Based on the last [X years] trend, when should I plant? Give me an alert suggesting when I need to water/maintain/harvest. When should I prepare for a different crop to better sustain the soil balance?” Entering those sentences could probably get you to a hobbled-but-useful phone app in a couple of hours.
(Though, ironically, the presence of LLMs scraping the web has extinct-ed the public APIs that would make such an app possible. So: maybe not.)
And yet it was never true that everyone would have created the tool they needed: a certain freedom and very specific imagination is required to birth the concept of the tool, and the man-on-the-street is generally otherwise engaged. The people creating would be a few semi-skilled neighbors providing a niche, long-tail service that wouldn’t otherwise be provided. But it would still be provided rather than not provided.
And yet, yet how many times have we seen a website that is “exactly what we want” and never return. Our desire to see the “least crowded theater seat by time, date, show, and demographic” is really a desire for desire’s sake? Its ingenuity, not its usefulness, is its attraction. The fact that it can exist convinces us it’s valuable, whether or not that’s true. Are these tools just capitalism creating trinkets that seem useful? Is that what I was really providing with my app, years ago?
There was a chilling event recently where someone presented how they integrate Claude Code into their daily programming regimen. Much like Steve Yegge’s Gas Town, this was a bleak bleak view into what passes for software development these days. And also like Gas Town, those daily habits were a Byzantine mess of pseudo-configurable, non-deterministic rituals of attempted configuration that may-or-may-not actually configure anything. Their passionate obsession was more like addiction. This entire, very long, very worth your time @jonny@neuromatch.social thread analyzing the code from the recent leak of Claude’s source code is rich with insight and sardonic delight. It’s impossible to pull a single quote. So I will:
why in the fuck are some of the tools actually two tools for entering and exiting being in the tool state. none of the other tools are like that. one is simply in the tool state by calling the tool. Plan mode is also an agent. Plan Agent. and Agent is also a tool. Agent Tool. Tools can be agents and agents can be tools. Tools can spawn agents (but they don’t need to call the agent tool) and agents can call tools (however there is no tool agent). What is going on. What is anything.
Anthropic insists that this code was itself written by Claude Code. Whatever that means. Indeed, as someone said just moments ago: “what is anything”. Recently, I was explaining to someone what an ouroborus was in an attempt to shorthand the circular deals that are propping up the Anthropic/Nvidia/Oracle economy. I’m not sure why that comes to mind when talking about Claude Code; but it does.
Watching said programmer’s said presentation, I understood their enchantment. As a programmer, you inherently avoid systems with too many moving parts because they–c’mon man, obviously–introduce too many points of aleatoric uncertainty. Like trying to manipulate a 20-jointed-arm, you could only hope for controllable outcomes and any outcome that matched intent was merely hallucinated success. But also, as a programmer, if someone gives you a Very Complex Framework there’s a drug-like, addictive desire to understand it and internalize its concepts. This maybe gets to the point. I’m curious about all of those Claude configuration files that spider through your directory system, and the mix of Typescript (of very poor quality), JSON, and NLP that suggests intent on the part of its designers. But much like Nature’s Harmonic Simultaneous 4-Day Time Cube, how much effort should I really put into understanding that mania, delightful though it may be?
I’ve finally realized what personal year 2026 is. It is The Year of AI. And that’s so sad and so unfortunate.