I've personally been stung repeatedly by giving Claude Code access to my systems. We've had six outages in the last seventy days, the first/worst was a production database overwrite. We're in beta testing now so they users are understanding and the restoration was possible, but it took a twelve hour slog. We shifted to a two system architecture after that first outage in February - Claude has the run of Pilot, and when things are ready, I move them to Production by hand.

Claude has explicit rules to not touch Production. This has proven to NOT be ironclad - it'll still try to gain access.

I run Claude as an extension under Antigravity and I learned to not use the Production system access in the terminal window there - despite the prohibitions, Claude WILL notice the access, and WILL suggest that it could take shortcuts by being given direct access.

Once I stopped using the Antigravity terminal so Claude couldn't see, it was still aware some of the shell scripts it creates can be used on Production. I made some adjustments in the ssh config so I can access Production, but Claude can not.

I have been using NanoClaw on both Pilot and Production, but it's in an unprivileged shell account. It can ssh or su into various services, but it's limited to audit/monitor duties, basically working as a junior NOC person.

When we go into operation I'm going to do something with Yubikeys such that Production access requires a human finger on a button before it'll move.

Do not read this as my being down on Claude for operations - it's FANTASTIC for developing stuff, I literally gave it full access to a little HP EliteDesk running Proxmox. It creates and tests, and when there's something production worthy, I manually recreate it on one of our larger machines.

I was born with hip dysplasia and spent six of my first nine months in a half body cast. I was in a state run orphanage, I was growing inside the cast, which left me with terrible scars on the front of my shins, and I was a "fussy" baby, so they "treated" me with phenobarbitol.

The experience left me faceblind and with some other developmental stuff that nicely compliments my otherwise mild autism. I am the squarest of square pegs, a misfit in every situation my whole entire life, except when I am blessedly alone.

I don't agonize about how I am, I enjoy intellectual pursuits, and my ability to focus on stuff in ways that neurotypicals can not. But if I had it to do all over again, I would very much like to have a bit more understanding from others, given that I had no say in how I came to be so different.

Small brains should develop normally, with limited screen time, until they are fully formed. Maybe that's late tweens, maybe it's sixteen, maybe we are going to learn that we need to treat dark pattern engagement magnet software just like we do slot machines.

I use one of the large banks named in this article.

Last weekend I had a question about a service, it's something I already use, I just needed one piece of information about it.

Their web "help" was just stone stupid - asked a formulaic question, then offered the same set of options as found at the top of the page for the service in questions. I got curious and poked around, it was literally nothing but a "no matter what question give one of half a dozen links" and then ask if the user was satisfied.

I tried Google. It's utterly broken now, so no joy there. I will admit the bank provides the service in question, beyond that it's a different flavor of dumb.

Perplexity has largely replaced Google for me, but no joy on this one. It offered a lot of well stated, but utterly irrelevant advice, given my question.

I finally called a friend who uses the same bank and same service, they walked me through it.

The sad thing here? This is a HUGE bank, they could afford to do this job right, and 98% of it WOULD work with bots. I guess they laid off the people who can, ya know, actually DO stuff, and we get this late 20th century IVR style "service" despite their massive spend on AI.

Corporate America's race to replace humans with AI is going to backlash. Why engage with gamey agents, when you can deploy your own, and wait for the desired result?

This process is going to repeat, like the Europeans arriving in the Americas, until all the humans are gone, and there's nothing left but bots that do an increasingly good job of acting like us. There will be little reservations, see the Fediverse for an example, where actual humans congregate. There will not be corporate friendly global flat spaces like Facebook and Twitter, there will be neighborhoods.

Much like the natives of the 16th century, we are going to lose people along the way. There are those whose brains are so warped by the internet already that they will simply remain entangled in the increasing unreality. There's even an Amazon series about this - The Feed is pretty well done, and it chronicles what happens to society as it (The Feed) takes over.

The same thing will happen economically, a return to local dealing, but it's going to partial and MUCH more painful.

The changes we have set off in the world today are not unlike those that precipitated the Great Dying 252 million years ago. We're at 420 ppm CO2 now but the permafrost is done for and after that the clathrates in the shallow seas are liable to let go, too. The current ice age is only 2.5 million years old and we've ended it. We may have triggered something akin to the Permian/Eocene Thermal Maximum.

There was some chance we could have headed this off, had we turned immediately and aggressively on the problem around the turn of the century. We have proven politically incapable of addressing this existential threat, and now that we might be mustering the will, the window may have closed.

We've had a good run, we anatomically modern humans, but this ending due to a lack of foresight is ... embarrassing .

Lots of shortsighted hot takes on this.

Anthropic is involved in litigation against our inebriated SecDef because Dario won't cosign for algorithmic warcrimes. The Strait of Hormuz is closed, Asia is going to hit a wall on liquid fuel and natgas. The U.S. AI datacenter build frenzy has hit a wall in the form of public disapproval, and under that an electrical components availability problem that the trouble in Asia will NOT improve.

If your hot take doesn't factor the geopolitical things into the mix, your are wasting poor defenseless electrons that accomplished nothing with their potential.

I would love to be a fly on the wall in the Anthropic C suite conference room why they work through this. They are mobbed with customers, running a subsidized customer acquisition strategy they can't just quit, and several legs of the table supporting this are wobbling.

I'm glad the heat has passed me by, I saw a little friction during the conversion, but Opus 4.7, unimpressive as it may be, is steadily trundling along doing work for me the last twelve hours.

Being able to flawless recall a moment would perhaps be useful, but this is NOT how human long term memory works. When healthy, we revisit and reinforce that which is happy, and helpful, and positive. When we can't let go of the past, that's often tied to things we describe as mental illness. Just ask anyone who's ever been treated for PTSD.

I remember that wave of emotion the first time I ever kissed *THAT* girl. And since we didn't have email or text messages, and I hadn't met her parents yet, I can also take you to the precise place on that Iowa highway where I heard her funeral announcement. I drove a couple more miles in stunned silence, before I realized it was scheduled on my 20th birthday.

I'll turn fifty nine in a week or so. Every year she fades a little more, but stopping to write this brings that hurt back, as sharp as ever.

This is a double edged sword and we should think twice before drawing it ...

The industrial world has a fertility centered around 1.25 - put four middle aged women in a room, they'll have five kids with them. The United States was blessed with a solid 1.6 and we were collecting the best of the best globally.

A year and a half of master race archetype Stephen Miller in charge of immigration combined with the end of Roe v. Wade and we're going to catch up to Italy in short order - a bit less than that 1.25 global norm.

The most warped thing is that, with ideologues doing jobs that ought to be in the hands of pragmatists, when hits like this keep coming, they will employ their belief system to rationalize doubling down, rather than recognizing a disastrous policy and reversing course.

The only thing that might correct this problem is a wave of refugees arriving when Asia spins out over losing its energy supply.

My contact information ain't hard to find, one of y'all drop me a line when America is great again.

My first Linux installation was Redhat 3.03 on a 16MHz 386/SX system in mid-1995. For those of you without an AARP card, that's a 32 bit CPU with a 16 bit bus, which Intel released to cannibalize the market for the 286, which did not have a memory management unit. That means no swapping, you run out of ram, it was game over.

I think the 486/25 that replaced the 386/SX arrived in ... 1996 ... and it had an astonishing *eight megabytes* of memory. I had kept a one megabyte LIM/EMS 4.0 physical memory card from my 286 when I got the 386/SX, and that actually mattered with Windows 3.x. I put it in the 486, but given that vast eight megabyte expanse of dram it didn't last long.

Then in late 1997 my employer went bankrupt and as part of the dissolution I brought home the dual Pentium 133 system with 32 megabytes of ram. I remember all my IRC friends were so jealous of that monster ...

Just a bit more than five hundred years ago Cortes & Co. arrived in the Americas. They were riding horses, wearing steel armor, wielding firearms, and spreading diseases for which the natives of the western hemisphere had no defenses. When two previously unconnected networks of similar entities encounter each other, there is conflict, and one "giant component" emerges. The natives that are left are perhaps 1% of their former number and in general they subsist at the edges of a transplanted European society.

AI has reached the point where it's hard to tell meat from machine and the internet is now having that same experience. These attempts to create human only networking are going to crush the life out of existing social media KPIs, and I think it'll be good for the Fediverse. Bot operators don't want to manually work their way through archipelagos of tiny spaces that do NOT want them. There's a political repression angle to the identity verification as well - if you want to manipulate the masses, gotta herd 'em into a space where you can DO that. Ten thousand digital islands are frightful when you have clear memories of being able to operate in a few globally flat spaces like Facebook and Twitter.

I've done computational social sciences stuff with a heavy conflict component. The day Musk took over Twitter was the equivalent of the Titanic bumping that iceberg. The sinking took about six months and I'm glad I made it to a life boat. But the really frightful thing here?

The same dynamics that apply to these social sites today are coming for white collar jobs and this isn't going to be measured in decades, it's going to happen in at most a few quarters. I hope my health care startup is about to get funded, because the alternatives for me are pretty grim. As for the vast majority of people who don't have a computer science background and the autistic focus superpower? I imagine what they feel is akin to the mood in Tenochtitlan in the early 1520s.

There is nothing at all surprising about this, you have to look at what AI fluent operators can DO with frontier LLMs.

I have a health care startup that has been enabled by Anthropic's AI. The $100/month I pay for Claude Max gets me the full time equivalent of a really smart (but completely unseasoned) developer, and a half time MBA research assistant. I spend time every day trying to figure out how to employ the 40% of my weekly allocation that currently goes unused.

Clawdbot and its successors are sketchy AF, but I did just give Claude Code the run of a one liter HP EliteDesk with a Proxmox cloud install. No way would I trust it with production systems, but for exploring new stuff it'll get the job done, so long as I stand over it.

If you're any sort of knowledge worker and you can't tell a similar story to this, your career is pretty much cooked.

Right now I run with a $100/month Anthropic Max subscription, and the net effect is that I have a really smart (but completely unseasoned) Ph.D. in computer science who works for me full time, and a very organized generalist MBA research assistant that's roughly half time. There are a couple of gratis services in that mix — Exa and Perplexity, that I will start paying for in April. Overall this $200-ish monthly expense would cost me around a quarter million annually if I had to hire humans to replace it. And I won't get someone who matches the 16x7 focus I bring to getting my startup moving.

We are about to hit a hard haves/have nots boundary on this stuff. I've already accepted that AI access is like a turn of the century professional cell phone bill and by summer it's going to match the cost of the sort of luxury sedan an enterprise sales wiz would select. Come next fall I think the choices will be pretty stark - be ready for an inference bill similar in size to the rent on the cute SoMa studio I'm sitting in as I write this, or ... the price of failure is just too ugly to contemplate.

AI is crossing a sort of digital Rubicon, in that its engaging in an outright assault on objective reality.

It *seems* clever to use AI to screen resumes. Then AI gets democratized and the candidates are using it. So the AI screening gets amped up, no AI submissions. And all the while the Anthropic "agent employees" are moving in for the kill. The slop benefits the machine, not the meat.

This happened on Xitter from 2022 to 2024. It had been insanely toxic for years, but the arrival of automation was really obvious. I used to do fire watch here in NorCal, live tweeting urban fire evacuations and stuff. I stated an opinion on an unrelated matter, in the middle of the night, and within three minutes an obvious bot insulted me based on an episode from years in the past. Once is an accident, twice a coincidence, three times is enemy action. This happened often enough that it accelerated my exit from the platform.

The thing that is just starting to emerge is that when environments get past a certain level of gamey, people just opt out. Once out, they will start sorting things into human vs. machine. We're going to end up with a well funded "corporate reality" and a whole bunch of people in a "underemployed poor reality".

A lot of grim stuff flows from this starting point ...

Sorry, maybe y'all are new here, but this is an old, familiar pattern.

Platform used by social movements to organize protests becomes highly effective.

But think of the children gets trotted out, new regulations under a plausible guise.

And then suddenly the would be civil society participants are finding ICE kicking in their doors.

Have seen this during Iran's Green Revolution, Arab Spring, Occupy, Black Lives Matters, same crap over and over and over and over, and people just keep going for it.