
"This method call is slow, make it faster" - Instead ask, how to not make this method call. Surprisingly often nobody thought of such a change and it turns out less work/code than the demanded optimization.

Functional program text author I am absolutely right

"This method call is slow, make it faster" - Instead ask, how to not make this method call. Surprisingly often nobody thought of such a change and it turns out less work/code than the demanded optimization.

Don't try to solve a problem before trying to remove the problem. A deleted line of code is worth ten times that of a new one.

LLM has accidentally solved being late. Simply set your alarm five and a half minutes later, as a "small allowance for lateness". This is a great example of something that sounds "linguistically plausible" but makes absolutely no sense. They do this in code generation all the time. And its not easy to spot, especially with rising complexity. In my experience, to spot such beauties, one needs the same amount of time one would need to write the code in the first place.
This company's worth $10B?

I write the program, because I read the program, because I run the program, because I trust the program, because I understand the program, because I mind the program, because I am the program and I am absolutely right.

Degenerates with generator (OpenAI, 2026)
This is GPT-6 Astra. Anything you can do on a computer, Astra can do for you. Fast.

Ex programmer is a completely different species than real programmer. Don't listen to them, they're mostly pretenders.

Its vibecoded lel
The Omarchy 4 branch is now 30,000 lines of new code. The majority of it was written by GPT5.5. It's been so, so good at QML. You still need to review, but there's just no way this scale of a conversion would be feasible without AI in a reasonable time. github.com/basecamp/omarchy/…

Truly amazing how well Deus Ex ages every year

A very late lesson I had to learn in my functional programming career is that putting a big state problem in a big state machine solves nothing by itself, all state issues are just recreated. This matters because I have worked on (and authored some) of the craziest state machines in Clojure. They suffer from all the typical problems, like stale reads, wrong execution order, etc., are just as horrible to debug, and sometimes make things worse by needlessly going async or diverting into async like execution flows ("effects") that are difficult to predict from reading the program text. What instead needs to be done first is eliminating as much state as possible, via infer/calculate view/layer on state, and all "effects" should ideally also be inferred by a reconciliation diff of the state against a real world snapshot, which needs to be kept maintained separately from the state. The actual state machine needs to be kept as small and compact as possible, returned effects should be avoided entirely. By itself, the state machine gives nothing regarding the elimination of state problems.

Relying on programs written by LLMs is dumber than relying on LLMs Attributing any level of trust to any LLM output should be liable.

People say rewrite but actually they port-castinate to another lang or framework to retain only half the features while making up for the other half with completely new bugs.

A text generator can not cheat. It can not copy, it also can't measure. Wrong tools lead to wrong results - no surprise here.

Just close your eyes and ask LLM whether slop is clean, and if not ask it to clean slop. Rinse and repeat, remember to keep eyes closed. Much faster than reading code. You are now a context engineer and have unlocked the future.

If you don't write the program text (code), you aren't programming.

How many repositories have you moved off GitHub today?

Because they want to sell you a subscription gated proprietary compiler (coupled with LLM and agentic workflow) behind an API next.