This doesn't surprise me at all. I'm constantly wrestling with the fact that Claude is confused by the Java setup on my machine and will waste untold number of tokens figuring out over and over how to run a test. I noticed and tried to fix it but there's no reason I had to do that.

Everyone complains about AI sycophancy, but I'm wondering who are these people whose friends and families are out here telling them that their ideas suck and they're wrong and they shouldn't do such-and-such? In my experience, most bad ideas go unchallenged because we all know we're not...

+1,000,000. If there is any constant that has held true over my career, it is that executives: a) never know when to stop, and b) always go way too far. Just as some companies doubled in size (looking at you, Meta) during the pandemic, this is so going to come back to bite them all in the a**.

The thing is, there's a whole crowd of frogs standing on the bank of the river and every time this f**king scorpion stings one of them, they fish the scorpion out of the river. Then he climbs right on the back of the next frog, who says "This time it's going to be different."

"One day you will come to a fork in the road and you’re going to have to make a decision [...] If you go that way you can be somebody. [...] Or you can go that way and you can do something [...] In life there is often a roll call. That’s when you will have to make a decision. To be or to do?"

6-3

My anecdotal impression from talking to folks is that companies are definitely on the hire and looking for people but are also being incredibly conservative and bouncing folks for minor imperfections. Maybe because layoffs make them paranoid about needing perfection, maybe "many fish in the sea"?

That number… can’t be true. Can it?

Suzanne Vega releases “Luka” 39 years ago this weekend. She said she once received a fan letter about it: “Dear Suzanne, Luka is the most compelling piece of music I’ve heard in a long time. There are no words 2 tell U all the things I feel when I hear it. I thank God 4 U, Prince”

This has always been a problem, AI has just made it 100x worse. Once you’re more than 2 or so levels away from the actual work, you lose all understanding of what that work entails and are prone to make decisions totally at odds with reality.

In an unrelated vein, been doing some writing lately and discovering that I just cannot seem to remember the first "c" in "syntactic." I just keep writing "syntatic." And keep getting corrected.

This fits with a hypothesis that I formed early on in the ChatGPT era, namely that we would reach a point where model trainers would start paying highly knowledgeable folks to just write things to be consumed only by the model. Now it seems it's not just writing, it's more basic work as well.

Shocking and sad. I don’t think I ever interacted directly with Soma but he ran DevDiv during some of my best years there. (I think my last memory of him was a ship party for some version of Visual Studio during an austerity period at MS where they staged the party, I kid not, on…

The first time around it was: let's hold on and we'll restore things once the four years is over. This time I don't know there's going to be anything left on the other end of this.

I fail to understand how there is anyone with a glancing knowledge of this man who would actually trust him in any way, shape or form. Just as with his former boss in the federal government, I would not do anything for this man unless he paid up front and in full. And even then I would not be sure.

On the one hand, good that this is being exposed. On the other hand, I would say I never trusted a charity marketed in the same manner as JG Wenthworth and your local “sue for damages” lawyers.

We've reached the point where "I've created my own agent harness" is like when folks used to say "I've created my own programming language." Good for you! But so has everybody else in the world.

I now have anxiety that sometimes I sound like an LLM. I was just reviewing a message I'm about to send, and I'm like, "That sounds *exactly* like the way Claude would put it." I'm pretty sure I don't have Stockholm Syndrome, maybe the models just digested too much of my online writing?

This reminds me of a story told to me by an old timer when I was a wee Microsoftie. Apparently there was a huge discussion around how Windows should indicate what button on a dialog had the focus (since mice were not as common and many folks used keyboards). One dev vociferously argued…

AI is a force multiplier, but I find these kinds of "experiments" to be annoying in two interrelated ways: 1. It has always, always, always been the case that "five really smart [cutting edge] people" can code rings around the average team of 40 or 50 people. OK, without AI...

When I was a student in New Haven, the tow trucks would circle the block on the main drag starting at 3:50pm, waiting for parking to be illegal at 4pm on the dot. As soon as the clock struck, they’d swoop in and start towing, no quarter given.

Southern son William Faulkner said it best: “The past isn’t dead; it isn’t even past.”

An exhausting aspect of agentic coding is the constant "problem -> request -> fix -> test" loop that's running. When I'm writing the code myself, I can often think through things and eliminate classes of problems at once, but the agent doesn't poke its head up and think, "I'm seeing a lot of...

An excellent time to tell a story I've probably told on here before (if so, sorry about that)... As a newbie on the VB team, I had to fix a bug in the math routines in OLE Automation. They were in assembly and I had some questions. I asked who I should talk to. "Oh, you need to talk to Tim...

I'm not saying it's not happening/going to happen, but JFC tech has been in a very long boom beyond even the COVID hypercharge and it is unsurprising that an extremely significant correction on tech employment was going to come.

What's the message going to be for folks who qualify but don't take it? (Speaking as someone who definitely would have qualified if I'd stayed.)

So in 2020, they had ~42,000 employees, by 2022, they had ~85,000 (+100%) employees, spent that year and the next year cutting 21,000 (-25%) of them, then over the next two years hire ~15,000 more employees (+25%), now announce they are cutting ~8000 employees (-10%). The drunkard's walk in action.

The thing you have to remember is in those days: a) autosave was not a thing, and b) crashing meant *the whole system*, so it’s reasonable that people lost significant amounts of work. (At least it was past the point where an improper shutdown could f-up your hard drive. I think.)

Deskilling as a byproduct of using AI should come as absolutely no surprise, as it’s the equivalent of going into management (i.e. having humans write the code for you). A massive chunk of my peers from when I started are very senior managers and unless they kept sharp on personal…

I feel like I need a T-shirt that reads, "No, I am not interested in adopting your f**king Claude skill." AI is definitely helping a thousand flowers bloom, but I am starting to feel buried under petals. I'll just figure it out myself. I am not going to clone your repo, attach your marketplace...

Someone remarked it’s hard to remember how bad people thought Reagan was (and he was!). But this was parody. Now it seems like reality.

I mean, you don't really need a translator

Everything else aside, my grandfather was a Methodist minister and I always thought the Methodists did church design well.

As a Gen Xer, when I called my 7 year old over to watch Artemis II take off, I did briefly consider the possibility that I was going to have to have THAT conversation with her and really hoped I didn't have to... Thankfully didn't.

This really feels like January 2020, with everything going on like normal as the storm clouds gather quickly and everybody acting like they aren’t going to get caught in the deluge.

As a writer, the worst thing about Rowling was her inability to constrain herself, something that got worse and worse as the books become more and more popular and she could clearly just ignore editors. The fact this is a miniseries not a movie just plays up that weakness so….

The problem is that building a UI framework, especially for something as old and large as Windows, is a generational project but most people ship v1.0, get their promotion, and then move on.

If ever completed, Graham Street would be the station closest to me but I’m not surprised and kinda doubt it will ever happen.

Honestly, consumer is so deeply not in Microsoft's DNA it's not even funny. And assistants in particular are Microsoft's Kryptonite, going all the way back to the misbegotten Microsoft Bob and Clippy (and his ilk).

something that's very hard for today's kids to understand: * there were only a few major brands of chewing gum * everyone knew all of them * each had its own banging anthem, whose words we still know * every one of these brands of gum fucks, to this day

It occured to me this morning that perhaps one reason why senior coders are taking to Claude more than junior coders is because it paradoxically takes them back to the simpler days when they were first learning to code. To wit: when the most senior coders were first cutting their teeth... 1/

"Is it Tuesday? It must be time for another Windows GUI framework." Every generation at MSFT thinks that *it* is the one that will finally create the one framework to rule them all. (Props to the ChatGPT diagram... I never did get to use "Avelon/WFF" or "Blacor Hybrid"...)

As I work with Claude, I increasingly realize how many things there are out there that I have thought about doing but which were just too frickin' tedious to bother with. Now that I can throw an instance at it, it's soon going to be the equivalent of the unread pile of books on the bedside table.

Man, I have been in SO MANY corporate meetings where someone got caught up short on a question they clearly did not have a good answer for and they tried to vamp their way through it. It's always completely obvious and it never works. Nobody is ever fooled.

My wife complains about the Temuization of nice clothing brands that she used to buy, how formerly luxe brands now all feel cheap and flimsy. As I start to get more and more documents at work obviously generated by ChatGPT and ilk, I can sympathize.

I feel like all the people who are like "...but the agents will handle the QA and bug fixing too!" have never called a large corporation's help line and gotten an LLM on the other end and experienced just how helpful they can be...

This will make filling out taxes easier! “Grok, I forgot what my SSN is, can you remind me? How about my wife and children’s’?”

I'll admit that, until now, I never understood what the Scroll Lock key was intended to do (nor do I believe I ever used it for any purpose whatsoever).

I, too, find the Slack AI bot to be shockingly helpful. My general mental model is that "ChatGPT/Gemini/Claude = helpful, vendor specific AI (looking at you MSFT) = completely useless" and it was a pleasant surprise.

This reminds me of my meeting with a prospective lawyer as my first marriage fell apart and me asking him how 50/50 custody works and him being kind of shocked I'd even be interested in that sort of thing and him explaining how most men wanted to be able to, like, move out of state.