Save as single file (.mhtml) isn't too awful in most cases.

Amazingly poorly. Things like floating frames/headers/footers render on each page. But the amount of scroll between pages doesn't take that into account. So the headers obscure actual bits of the main frame/article on subsequent pages.

Do you mean scribes? Because I'm pretty sure all other writing jobs involve working with machines.

Oddly, journalism doesn't exist to provide jobs to people who wanted to get writing degrees. It exists to report the news.

The loss of classified advertising and the rise of the internet have drastically reduced funding for traditional journalism. The shift from daily newspapers and the evening news to a 24/7 news cycle have also cut the amount of time journalists have to gather and vet information. If you want to take notes with a pad and pencil and have your photographer take their film to a darkroom, you're simply too inefficient in the 21st century.

The critical missing pieces are editors. In the local paper I subscribe to (San Jose Mercury News), I see obvious grammatical and spelling issues on a daily basis, which implies nobody but the original journalist even read the article before it got published in a print newspaper. And they didn't bother to use a spell-checker, either. That extends to the reporting, too; studies are misquoted, statistics misused, important facts left out, and assertions unsupported.

That's before the use of AI to write drafts. So I don't see how this makes it much worse.

In fact, AI might actually make it better, whether it's writing the draft so the journalist can take on the role of editor, or taking the role of editor after the journalist writes their own draft. Because clearly none of that is happening right now at my local paper. Doesn't matter if it's always right. In fact, it's probably better if it hallucinates 10% of the time, because that way the journalist can't rely too heavily on it.

(Yes, in an ideal world, we'd have real human journalists and editors, and enough time for vetting to take place. Are people willing to pay for that? Evidently not.)

That's all fine and good if you have connectivity when you're scrolling.

When I'm getting ready to fly, I preload about 50 tabs in my browser so I have something to read. It is a horrible PITA to hand-scroll each tab down to the bottom so all its images load.

It's even worse when the browser crashes. Because then I get to do that again, for every tab. Usually on crappy airport or hotel WiFi.

Sure, load all the images after the text. But please do load them all, rather than waiting until I scroll.

And it's the exact same stuff they said about factory automation.

And it's the exact same stuff they said about using machines in factories at all.

Yes, companies that offshored too much or stuff that was too important to the core of the company did screw themselves over. But most companies did not go that far and lose their best employees and know-how. That would have caused most companies to collapse.

And that's where we are now. Companies that use AI for too much or stuff that's too important will discover they've lost their core competency and knowledge. Companies that use it in more limited ways will have a competitive advantage over companies that don't use it at all. And even down at the bottom, engineers who figure out what it's good for will be more productive than ones who don't. Not 10 times as efficient, but 10% more efficient is still enough to matter.

I've been writing software for over 30 years. I used to spend 20% of my time writing and maintaining unit tests, one-off scripts, and stuff like that. I would not trust AI to touch my core code. But tests and one-off scripts? Stuff that fits easily in its context window? Sure.

No, it doesn't *always* work, but it *usually* does. And it doesn't take very much time to figure out when it's not working. And rather frighteningly, it does a *better* job on most of those one-off scripts than I would. (I'm not going to spend a bunch of time on error handling and comments for code I only need to run once. But the AI does.) Sure, I could have an intern or junior SWE write the same tests or scripts, but that'd take hours, vs. minutes.

If 80% of your engineers are able to leverage AI to do the same work in 80% of the time, it doesn't matter if the other 20% can't figure out how to make AI work. You just lay them off; the other 80% are efficient enough to compensate for the layoffs. And that's one of the big things causing stress right now. You either figure out how to use AI, or you'd better start learning plumbing or something.

It also sucks if you're a new-grad SWE. Because the AI is better and faster than you *are right now*, and a lot cheaper. If you had a couple years experience, you'd be in the first pile of engineers trying to be in the 80% who keep their jobs. But you aren't. And you can't get that experience, because nobody will hire you until you already have it. The best you can do is work on open source or personal projects to build experience in the evening, after you bag groceries during the day. And hope that you get better faster than AI does. (Seriously, that's what my son is doing.)

Unfortunately, I don't see a way off this unamusement park ride. Even at a 20% gain, any individual or company who doesn't use AI is going to lose out in speed or cost to someone who does.

(Actually amusingly, the companies that try to push *too much* AI won't end up any better than the ones who push *too little*. Those are the ones who will vibe code themselves into corners, or AI will break stuff in prod, and nobody will know how to fix it because they laid off too many people too quickly.)

And that is: The register must honor the lowest price displayed in the last 6 hours.

There's nothing wrong with a store dynamically lowering prices. We just don't want them to go up while we're shopping.

Virtually all shoppers are wandering around with smartphones with cameras, so it's easy to take photos of the tags. If you catch the register not following the rules, you get some payback. (What is an implementation detail. Double the difference? One item free? Depends on how frequently that store has a problem?)

I already take photos of every sale tag when I'm shopping, just so I can confirm I'm getting the sale price. It also saves a ton of time when an item fails to scan at all; rather than them sending a runner back, I just show them the photo of the shelf and they key in the price.

Between 2012-2021, San Francisco issued a total of 11 concealed carry permits. (source: https://calmatters.org/politic...)

In 2022 alone, San Francisco police seized over 1000 illegally-possessed guns. (source: https://sfbos.org/sites/defaul...)

Many people in SF are carrying. Few of them are doing so legally.

Why do you think there is any overlap between the SF Board of Supervisors' demands and reality?

This is the same city that keeps trying to ban corporate shuttles so that all the tech workers will have no choice but to ride public transit, and then will have no choice but to pay to fix the transit system for everyone else too.

Full scenario:
- SF Man attacks Waymo
- Passengers call 911
- Police never come
- Man finally breaks through window and attacks passengers
- Passengers open fire on the man in self-defense
- Police finally show up and arrest the passengers
- Man sues Waymo for medical expenses for his cut hand and gunshot wound
- Public blames Waymo for (1) inciting the incident and (2) not driving off to de-escalate
- SF Board of Supervisors demands Waymo change its terms of service to forbid armed passengers

Google includes actual headlines. News sites complain and demand payment.

Google stops linking to sites. News sites complain and demand payment.

Google links to sites using paraphrased headlines. News sites complain. (I'm sure the demand for payment will follow.)

I worked for many years at a Big Tech Company. SWEs and UX engineers were promoted for launching Shiny New Features, not for improving performance or fixing bugs in existing features.

Sure, every once in a while the company would announce some initiative to fix existing stuff. But it never helped anyone I knew get promoted.

I will believe this new initiative from MS when I hear from MS engineers that *they* are actually being rewarded for fixing this stuff.

See this slashdot article from a year ago: https://slashdot.org/story/25/...

If you configure the tool to minimize delusional thinking, it does.

Of course, if you configure the tool to maximize engagement, well...

It's easy, if you sort them first before compression. I have roughly 100 trillion of each number 0-9, and a single decimal point.

There's rarely a right way to write code, though there are definitely plenty of wrong ways. :)

Every SWE goes through phase 1, which is writing yourself into a corner by not thinking about maintainability and interfaces. And then phase 2, where you write such a complex interface that it collapses under its own weight. Only then do you get to the point where you're even aware of the need to balance interface complexity and upgradeability vs. code velocity.

Usually at that point you mock some stuff up quickly to see where the pain points are, and then go back and do a more maintainable design. And try to convince the manager that the mockup is *not* production-worthy code, because it's full of shortcuts and empty of tests and light on error handling.

Even if you're coding something the right way for the *current* requirements, it may not be the right way when requirements change. Which they do.

But if you're writing code for money, there are finite resources. So you never get to do everything the way you want. Sometimes you skimp on error handling. Sometimes you skimp on tests. Sometimes you tunnel holes in your API to get at that one state variable that just isn't cleanly exposed and yet absolutely critical to patching some other bug. All those things add risk. And at some point, you need to mitigate at least *some* of that risk, hopefully *before* it costs you downtime / money / customers.

It's always a hard sell to management.

And it's often a hard sell even to the engineers. If you want to get promoted, you're far better off writing a spiffy new feature rather than decreasing technical debt.

We've had decades to write poorly-tested poorly-reviewed code. But that's ok; as long as it kinda worked, we insisted on shipping it.

AI is now good enough to show what I told my managers for years: That technical debt builds up, and at some point the bill will come due.

The bill is now due.

Thanks to AI, it's now easy to find bugs. And relatively easy to confirm they're exploitable. But thanks to all the rest of the technical debt, much harder to fix the bugs. AI isn't good enough to fix the bugs yet, either, at least not without creating new ones just as fast. So it's a target-rich environment for hackers.

I'd say "I told you so", but I got out of that rat race a few years ago.