At the airport, on my way to New York to meet the rest of the Amp team, so no big intro this week. If you’re around, you should come meet us.
The Next Feature Fallacy: “For people who love to build product, when something’s not working, it’s tempting to simply build more product. It leads to the launch-fail-relaunch cycle that I alluded to in a previous essay, Mobile app startups are failing like it’s 1999. However, this rarely works, and when you look at the metrics, it’s obvious why.”
So, Google released Gemini 4 Argon. There’s a whole bunch of numbers and percentages in that post and most of them I don’t care about or trust, but this paragraph made me purse my lips as if to let out an impressed oooh: “A team of Argon agents analyzed fleet-wide profiling telemetry to autonomously identify and apply memory optimizations across Google’s data centers, freeing up over 300 TiB of memory once rolled out, with an estimated 500 TiB to 1 PiB in total savings.”
Daniel Lemire with a fascinating perspective: “There is a divergence in society. Some of us have been frustrated by the lack of progress. We don't like what has been happening to our civilization since 1970. We expected lunar bases, rejuvenation cures, cancer cures, space trips... and all we got was Twitter. […] I feel that the people who love stagnation got enough of what they wanted and I hope that we are at an inflexion point where we finally break free from the oppressive rule of stagnation.” I’ve had similar thoughts before: if you walk through this small town in which I live, nearly everything looks like it did thirty years ago. One reason for that, of course, is that we now all spend many hours per day staring at screens and those screens contain large portions of our lives and no one has their screens sitting on the porch (although they sit with their screens there). Reading Lemire’s post made me realize, again: for most of us the most positive vision of the future looks like the 90s. And that’s… sad, isn’t it?
“Several vulnerabilities have been discovered in the Linux kernel that
may lead to a privilege escalation, denial of service or information
leaks.” Several, in this case, means 1597. Sign of the times.I haven’t read more than the first paragraph yet, but in a few seconds, once you’ve read it too, you’ll agree that, yes, one can stop there and it would still be enough: “One warm spring night in 2011, a young man named Travis Hughes stood on the back deck of the Alpha Tau Omega fraternity house at Marshall University, in West Virginia, and was struck by what seemed to him—under the influence of powerful inebriants, not least among them the clear ether of youth itself—to be an excellent idea: he would shove a bottle rocket up his ass and blast it into the sweet night air. And perhaps it was an excellent idea. What was not an excellent idea, however, was to misjudge the relative tightness of a 20-year-old sphincter and the propulsive reliability of a 20-cent bottle rocket. What followed ignition was not the bright report of a successful blastoff, but the muffled thud of fire in the hole.” But still, I want to read the whole thing now.
This was an amazing peek into a community I didn’t even know existed: Pirating the Pirates.
So, rumor is that if you keep saying “there’ll be a rug pull any time now, the prices will go up” they will roll you up in a rug, because look at this: “we found opus 5.5 to be 2X as efficient as gpt-6-sol in our internal cfo.ai evals so i was not expecting 6.1-sol to improve on that much but to see it 3X the efficiency of opus 5.5 and 2.5X against even sonnet 5.5 is actually [mindblowing]” GPT-6.1 Sol cost $2.72 vs. Sonnet 5.5’s $7.60 on the same task, with the same score.
Anthropic on GLM-5.3: “Here, full credit is awarded for a full control-flow hijack. We evaluate several models on 100 tasks from the benchmark (selected at random), and find that GLM-5.3 develops full control flow hijacks in 4% of the trials; Claude Mythos Preview did so in 6%. Although GLM-5.3 performs below Claude Mythos Preview here, a meaningful threshold has clearly been crossed: earlier models, like Claude Opus 4.6 and GLM-5.2, do not succeed in any of them.” That’s wild, isn’t it? Mythos, the model so dangerous that Anthropic only wants to give it to its highest-paying customers, got 6% and GLM-5.3, the Chinese model, gets 4%. And they managed to disable “GLM-5.3’s safeguards between 64% and 100% of the time with simple techniques in our simulated tests.” What I wonder now: doesn’t that mean that Pandora’s Box has been opened and that no “pacing” will help, that whatever Anthropic campaigned for when they got Mythos locked up by the U.S. government, is now in the past?
This is amazing and motivating and inspiring and I want every company and every organization and every government in the world to do this: “by the end of the ZBR cycle on July 1, 2026, Idaho had eliminated roughly half (49.1%) of its regulatory code. Regulatory volume fell in each of the eight annual code publications from 2019 to 2026 under ZBR. What Idaho has done, other states can achieve as well.” ZBR refers to Zero-Based Regulation and that “borrows its central premise from zero-based budgeting: last year’s baseline should not be presumed to be the right starting point. Traditional regulatory review generally asks what should be changed or repealed, if anything, from the existing code. ZBR begins with a more fundamental question: If the state were starting from scratch today, which rules would it adopt and in what form? The practical effect is to reverse regulatory inertia. Instead of requiring reformers to prove that an existing rule should be eliminated, the agency must justify why the rule should continue to exist.”
Destroy Any Website. Awesome. More.
There have been many variations of the original motherfuckingwebsite.com and they never quite topped it, did they? But this newest one, the Modern Motherfucking Website, is quite good. I had Amp change my personal website to use that styling and… I think I might just push that up?
Jeffrey Katzenberg, co-founder and CEO of DreamWorks and, prior to that, CEO of Disney, on AI and Creativity and filmmaking: “The Disney archives held astonishing recordings of Walt explaining his creative process. His own writings. His notes and storyboards. Work product captured at every stage of his process. This was truly a gift. Listening, reading, sitting with the work itself, we heard him talk about character, about emotion, about how an audience feels when a character truly comes alive. He talked about making bold choices and refining a scene until it genuinely moved people. We didn't hear a word about pencils or paintbrushes. In fact, Walt was famous for being a technologist, forever hunting for state-of-the-art tools, often inventing them himself to achieve the images he saw in his head. But he never defined animation by the tools. He defined it by whether the audience believed the character. His principles were timeless. The tools were not. […] Tools are never the point. The instruments change with every generation. What endures is taste and imagination. The magical ability to make an audience feel. One of the greatest storytellers of our generation, George Lucas, succinctly captured the eternal essence of this issue: ‘It’s not the how, it’s the why.’” The whole thing is very good.
Is Mathematics Over, or Just Graduating?: “Once airplanes get cheap, we will also discover satellites. We will map the whole terrain for mineral deposits and other valuable resources, and then extract them. It will be messy and dirty, and today’s explorers will be deeply appalled at what it does to their beautiful jungle. They will long for the good old days. They will say this is no longer exploration but a commercial undertaking, large corporations wrecking the area they were so keen to explore. They will be right, but it won’t change much, and large companies will find unimaginable riches there. […] Do not bet against the airplane. The skeptics who say AI can’t conjecture, can’t theorize, and can’t explain are describing last spring’s models. Within months, they will be describing nobody. The evidence is already on the table, and those with eyes can see it. Every “it can’t do X” has a short shelf life now, and anyone building their identity on one of those sentences is building on sand.”
Wilhelm is building a personal super app and it sounds like such a good idea. By now we’ve probably all experienced how easy it is to add more functionality once you have the skeleton of an app and deployment set up. This leans into that.
I’m not in need of one, but there are new Kindles and, man, they look good.
Jason Fried built his own writing application and there are a thousand things I could say here, but instead let me just say this: watch the demo video and think about how you would’ve built all of that three years ago. Also: god that guy is great at demos.
So there’s been a bit of a kerfuffle because Figma doesn’t allow any arbitrary MCP client to connect to their MCP server and the majority opinion seems to be that Figma should just allow that. Okay. But to me the most interesting thing that came out of the discussion was this post by Wes Bos and his reply further down: “Figma doesn’t want you to use MCP. Amazon doesn’t want you to shop with agents. Reddit won’t let Claude access threads. X won’t let ChatGPT read tweets. It’s starting. This happened with APIs 8 years ago. […] I’d be willing to bet Amazon makes more money when you use the app. Grocery stores make more money when you walk the cash wrap. Airlines make more money when they scare you into buying insurance. Dark design patterns and consumer usage data make money. They don’t want to give that up for some chat bot deciding what the cheapest red shoes are.” It seems so obvious when put like that, doesn’t it? He’s exactly right of course: I remember many many years ago there were conference talks and excited conversations about how once everything is REST and everyone uses OAuth we can just pipe this into that and build our own client for that and have our cake and eat it too. Except that didn’t happen. Not even close. So why would all the companies now allow anyone to bring their agent?
Jurassic Generation. I want to re-watch Jurassic Park now.
That being said: I think Sign in with ChatGPT might be one of the most interesting things launched this week. (Amp was a launch partner!)
Remember the Rippling vs. Deel corporate espionage drama? This isn’t as good, but it’s close: Factory kicked out a board member because (that’s the accusation) he shared confidential information with Factory’s competitor Cognition. But then the accused said he actually quit. But Keith Rabois said even that would be unethical and a lot of dirt was thrown. Very entertaining but I’m willing to bet that in a week no one will even talk about it anymore.
Made by mechanical means, by Felix Rieseberg, on how he used Claude to rebuild his personal website. What would you call this? Programming? Designing? Or just… building software?
Werner Herzog: “I only tell everyone out there: it will be a counterpoint against what you normally learn in film schools. Young people spend way too much time in film schools. It’s way too expensive, way too long. And I don’t like the kind of teaching there. So I said, ‘Two things I really will teach you. Number one: lock picking. And number two: forging of documents,’ like forging a shooting permit. And a film like Fitzcarraldo could not have been made without massive forgery, so you have to have the right criminal energy and the skills to learn.”
Stewart Copeland playing the drums. YouTube says the video’s been there for 19 years. I think I’ve watched it at least once in each one of those years.
The grocery-store comparison makes the MCP discussion easier to picture. A direct answer can be convenient for the customer while removing the detours a company earns money from. Whether an agent can connect and whether the business wants it to are quite different questions.