You missed the point.
A friend of mine made a tea-timer app for the iPhone. Where you enter what tea you are making and it sets the correct number of minutes. Today, people would ask ChatGPT before they go on the App Store to search for something like that. That's the point. Not what YOU would do, but what the general public does. And for Jane Doe, ChatGPT has replaced a lot of these specialty apps.
Skills are great until they aren't. The problem is exactly what you say: You can literally create a skill by telling the LLM to create a skill. Which is... why the f do I need to make a skill if the AI already knows how to do it?
Properly created skills, that include reference material, specific instructions, and expert knowledge the LLM doesn't have or that's buried somewhere under 10 layers of reddit posts is genuinely useful. But most skills aren't. I'm constantly underwhelmed by hyped-up skills that, when I look at them, turn out to be bland, basic instructions.
"Wear-out failures aren't always different from infant mortality failures, though."
That are completely different issues, always different.
Only in the most utterly pedantic sense of the word "different" wherein you mean that by definition, infant mortality occurs early and wear occurs later. But they can both stem from the same root cause, which is a component not built with adequate tolerances. In other words, they can be the same fundamental issue, just expressing themselves at different times.
That would be reflected in a low MTBF rather than the example "MTBF of 600000 hours".
Most products don't have an MTBF. And things that do have MTBF are based on accelerated wear testing on a small number of units, and can have almost zero basis in reality. There was one model of major brand hard drive about twenty years back where a couple of RAIDs where I used to work got lost because the drives all died after such a predictable number of powered-on hours that by the time you could re-stripe the RAID, the next drive would die. They ended up making sure the drives had a high variance in powered-on hours before using them in RAIDs after that.
Do you really understand what that means? Because it seems clear you don't.
It's an example that proves or demonstrates a point. And the point it was proving was that the same root cause can be responsible for both catastrophic early product failures and premature deterioration-related failures.
I'm skeptical of this.
People who don't understand that wear out failures are a different statistical mechanism to infant mortality failures I'm sure are very sceptical. I guess you also don't believe any HDD can fail within the first month of operation because they have an MTBF of 600000 hours?
Devices get shipped with manufacturing defects and bad batches all the time. That has zero to do with wear out failure rate.
Wear-out failures aren't always different from infant mortality failures, though. If the defect is caused by a design flaw in a component resulting in a structural weak point, that same weak point could fail after one use or a thousand, and it could just be a ticking time bomb. Someone could get very unlucky and find one that is slightly weaker, and it could break much sooner.
Case in point, I recall a machine where some people would have to have their laptops replaced every couple of months for motherboard failures, while others lasted for years, and it turned out to be dependent on how people closed their laptops at the end of the day and whether they pressed on the bottom of the case in a particular spot that wiggled the GPU. But if you handed one that had been working for five years to someone who used their left hand under the laptop while closing it, it would be dead in two or three months. If you replaced it with a brand new one, the brand new one would be dead in two or three months. And so on.
But again that's backwards. I let my AI coding assistant write tests exactly so that I have a reproducible, deterministic ground truth that I can inspect for logic errors. If you let the AI do the testing directly, it's a toss-up if today it feels like it or it decides that your desired result is "everything green" so it'll just give you that answer without actually testing anything.
LLMs obviously aren't human. But that's not the point. The point is that you don't need to know a programming language to program a computer.
For example: Especially in the early days of smartphones, there were mountains of small tools that did something that's essentially a simple script. Egg timers, remind-me-tomorrow apps, tons of stuff that's basically a wrapper around a web search, and so on. Almost all of those are today something ChatGPT can do, and are easy enough to describe.
and if you don't approach the LLM wit a borderline adversarial level of suspition and mistrust
I've actually outsourced the first iteration of that to AI.
At the end of coding sessions, I generall ask it to spin up an adversial agent whose task it is to find flaws, inconsistencies, mistakes, bugs, etc.
I'd still strongly recommend to anyone shipping AI-generated code to have a thorough human code review, but for a first pass at surface level, the AI can check itself if it has some kind of separation between the generator and the reviewer.
Personally, after some experimenting, I've concluded that:
What I use it for these days is rapid prototyping. Using AI I can make a functioning prototype of an idea I have in a few days instead of a few weeks. I can quickly test out an idea for a feature or change in my existing codebase.
What I also have been using it for a while is as a code-reviewer and for someone like me who is mostly working solo, that's an insanely valuable thing. Not just the bugs themselves, but also the ability to discuss code with someone. Basically, AI is a rubber ducky that replies.
Mickeysoft lost the plot again.
The game-changing element of LLMs and agents is that they have a non-computer-language interface - you can talk to them in plain English (or Chinese or whatever). This makes complex computer interactions available to a much wider audience than any previous computing invention. Adapting humans to the AI is getting that backwards. It's a sign you don't understand the first thing about why AI is so widely successful.
I predict this will be an abject failure and a year from now will quietly be pushed to the very back of the bottom-most drawer, never to be spoken off again (though it is getting crowded in there for M$).
Meanwhile, in the real world, everyone I've talked to about these things already has a skills file to teach the AI about their personal coding style. This is what actual humans came up with, independently several times, and it works: Adapt the AI to you. Plus we've been writing coding style guides for humans for a long time now, we've got plenty of experience doing that at least somewhat right.
If you ask a bare LLM to do stuff you'll get a mess. I run Claude code with a multipart harness that compels the model to treat the code as the mathematical thing it is, not human prose. LSP Enforcement Kit keeps low capability tools like grep out of the mix, Abstract Syntax Tree things handle the code as its meant to be understood, so that's CodeSight, OptiVault, Serena. About a year ago I gave Antigravity the boot after their awful 2.0 release and went back to JetBrains IDEs, which now offer code structure wise MCP services. Code review with the Ponytail skill chases away LLM excesses, and there's a smattering of other stuff like that in the mix. And then you hunt for whatever additional guardrails you can get the LLM to apply systemically.
I've evolved my own higher level project methods and they're comparable in a lot of ways to what Matt Pocock offers, but his stuff is more structured, more carefully considered, and I have to work with others now, after a long time solo. I'm taking his stuff, implementing a Kanban/wiki system, Gitlab CE, and NanoClaw for anyone who wants agents. I expect quite the slog between now and New Years, but in 2027 we're gonna stop trudging and start gliding over the landscape.
Because it embeds into html, rather than rendering the html. It makes for very fragmented code.
In my experience, most sufficiently advanced use of PHP involves rendering the HTML. You can do it either way.
And most sufficiently complex PHP uses PHP as a full-fledged language, complete with classes, etc., and to the extent templates are used, the code in them consists of tiny partial-line-long fragments of code that call some method in some other file that is basically just code.
In other words, very fragmented code isn't caused by PHP; it's caused by the way that the developer chose to use PHP.
I fly 200ish flights a year so figure 70-80 nights away.
2025 was 100% hotels but 2019 was about 80% Airbnbs.
Airbnb will never get my business back. Ever. The professional scammer hosts are so good at finding a way to nail me for a fee. Iâ(TM)ve taken perfect check out video and still had to spend hours fighting fees.
Airbnb allowed making guests do it.
They lost so much money to hotels theyâ(TM)re banning hosts from requiring it.
Yepp, as I said: The system is built first and foremost to defend itself.
Personally, I don't think that police should have any right whatsoever to enter your digital devices. Yes, including for terrorism and kiddie porn. And I am 100% for us nerds developing defenses against this overreach. This case is a lesson that teaches us distress PINs are a good thing, but we need to improve on them, make it less obvious they were used, etc.
Prediction is very difficult, especially of the future. - Niels Bohr
