Some thoughts on LLMs

11 June 2026 · 3 min · 448 words · opinion, llms

All photos are shot on an Olympus Trip MD3, loaded with an Ilfords HP5 Plus, at Goulinas .

The world of software engineering is shifting really fast. The stochastic parrot that was good for sloppy snippets is now smarter and can code really well. Not smarter in the sense of a human. Smarter in the sense of “this multi-dimension vector really knows what comes next”.

It has been called the third golden age of software engineering* and really smart people think this is the new way forward. The problem was never writing the code. It’s all the processes that lead up to that moment, but also, the decisions taken when writing the code. LLMs are just another arrow in the quiver.

The ethics are bad. Really bad. Copyright infringements with multiple lawsuits already filed, high resource consumption that leads to resource deprivation for local communities, and use in war such as the attack at the Minab school at Iran. The points mentioned here only scrapes the surface.

I am not surprised that multi-billion corporations are acting like that. What I am surprised from, is the complete lack of regulation.

I wont pretend to understand the economics behind the bubble but a great resource in everyday English is Ed Zitron’s blog and podcast . If you just want the numbers check Is AI Profitable Yet? .

In sort, no one is profitable and everyone is starting to run out of money. Claude burns somewhere around $8 to $13 per dollar earned on subscription plans . GitHub has already transitioned to usage-based billing. Others will soon follow.

The bubble will inevitably bust. The magnificent seven own about 35% of the US stock market and that will not fare well for, you’ve guessed it, me and you.

What does the future hold?

LLMs are definitely not going away. The BoochBurgers are not exploding as much now and newer models are more capable than they were mere months ago. Software engineering is going to change; much like it did when higher level languages were introduced. I can also imagine a lot of useful applications, especially in the a11y space. Take for example a screenreader that can infer what the screen is showing without actually having to “read”.

On the other hand, in this era of technofeudalism, companies are using AI to undermine working conditions and fire workers to maximize their profits. Governments are already using it to surveil and kill people. Apps are being enshittified and “AI” is being shoved at every nook and cranny. I trust that we, humans, will make the worst possible decisions going ahead.

* Substack hosts Nazi content . Use something else and let the authors know.