"And when it does, will we be grateful for the places that this delirious moment took us? Or will we wonder where in the world we thought we were going?" I think about this every freaking day.
I think an increasing number of people are asking What’s it all about? What’s the point? Am I happy doing what I do? I think asking the question(s) is valuable and important. The answers are hard. Hell, even finding a place to stand while you ask is hard. This morning in the big Venice fish market I saw 4 seagulls fighting loudly and with great energy over a piece of fish guts. I’m worried that there’s a metaphor there.
there was a whole lot of Discourse about this this week after this whole post (https://x.com/deedydas/status/2055491938464489888), and the seagull analogy perhaps seems better than I wish it were.
We've reached the peak of building for the sake of building.
It was always a problem with the tech industry, and most people working in it, that the emphasis was on the HOW instead of the WHY. Many people have been yelling into the void about "business value" and "solving real problems" for years, and for a while I thought the message might finally start to resonate post-ZIRP era... But no, with LLMs we just completely discarded outcome as a goal and went for maximal output. Or, dear god, for maximal *input* with all the tokenmaxxing talk!
Maybe the eventual turn of the currently exponential hype curve into the usual S-curve will get us back on track. At least I hope so!
It's sort of odd, because pre-AI there was a compulsion to build, but it was a compulsion to build companies. Now, it's more of a compulsion to build *stuff*, which is probably less cynical (there's more genuine fun in that; it certainly could be valuable; etc). But, it still seems rooted in a similar motivation.
I just had to come here to say that ever since reading this piece I think "this is the MOH-ST words" every time Claude writes a 20-page treatise to answer a yes or no question
So much of it all has been pointless. Unfathomable amounts of wealth wasted for what?
The smartphone at least had immediate obvious implications. It's turned out to be a profoundly terrible invention for society, as it led to the rise of endless social media and all the distortions that has created - but that wasn't obvious immediately, and it has had plenty of actually useful things to at least slightly counterbalance that.
But AI? Nobody has built anything with AI that has truly been Good. We're 3.5 years in and the 'best' we've gotten is using it as fancy autocomplete, or maybe as a tutor to learn stuff - but part of learning is often the struggling to find good info as it helps your brain etch it deeper. All that time, wealth, and talent squandered.
The bubble grows bigger; at some point it must pp. And then we all suffer for their vanity.
I don't think I'm quite so pessimistic about it, that nothing good will come of it, though for the moment, it does sort of seem to be hard to see. Like, it definitely "works," in the sense that it can help people accomplish stuff a lot faster than before, but is that stuff useful? Ehhh? Maybe? My hope, I suppose, is that it goes in the opposite direction of things like phones and social media, where they seemed good at first and then created problems; maybe this one will be somewhat messy at first, and we all eventually figure out how to make it do very useful things that really does make lives better. But I'm not sure vibe coding apps is the thing, just yet.
If you are good at what you do, it's a multiplier. I can draft emails way faster, or auto-customize resumes, that kind of thing (Though so can everyone else, so the arms race gets worse). And for me, that can be helpful - but then I'm not doing the work myself, and I suspect that leads to my own skill development atrophying.
I've seen some limited creative use too - someone who REALLY WORKS at AI art can make good stuff; the Monoverse on Youtube is proof enough of that for me. But you still need taste and a vision. And it's a ton of work.
But if you lack that? You churn out slop. You're just an avatar of the AI, really. And that seems way more common.
Vibe coding apps is definitely not it. What is it requires that grit, the determination to work hard, the caring about the end product...and that lack of care is already so widespread. I'd argue that lack of care is why the economy is rotting; why politicians are so much less responsive; and so much else.
If AI were to help us, the first problem it needs to be tackling is 'How do we make more humans give a shit?' because that's the crippling issue we need to overcome.
And when there's a (apparent?) gold rush, it seems to attract a lot more of that too. That's the most striking thing about that whole Corgi insurance story - not all the stuff about how many hours they work or whatever, but the fact that they're putting in those hours to solve a problem that there's no way in the world they actually care about, other than in the belief that it could produce a big business.
Exactly. There's a ton of 'let's do X, but with AI!!' right now. And I mean, if you really spot a space to innovate? That's fine. But most of these aren't that. They're just attempts to cash in - either on investor money to ride high for a year or two, or even better, a buyout/IPO cause the odds any will be truly viable products are...slimmer. And that may be optimistic.
Video game producers figured this out a long time ago: give people a capacity/"energy" meter, and they won't stop playing when they run out of capacity, but rather they will stay glued to the screen waiting for more, or will even pay real money to get more -- *and they will stay up late at night to make sure that they consume all of their capacity before going to sleep*. And then guess what's the first thing they will do in the morning?
Anything that creates a sense of urgency is good for business (even if it is completely made up).
One more pull.... just one more... I have to, or else I will fall behind!! Time is running out! I must use all the tokens! I can't let any go to waste! There are so many ways to justify it.
AI agents are just the latest game.
But if you were playing tennis, would you have to hit the ball just one more time? Why? What difference would it make?
I've played baseball for a long time, and whenever you're taking batting practice, everyone keeps saying "one more" for about six swings, until you hit one you want to end on. It feels like there's some of that here too, where you don't want to stop until you end at a good stopping point. But everything you add creates a new thing to be done, so that there's constantly one more thing to do before you stop.
The "blurry JPEG of original ambition" is the right framing. Output volume measured in lines-added-to-folder decouples from the actual product question, and the loop where each setback resolves with "I'll rebuild with Opus 4.8" is the same admission at finer time scales. The prompt-mile and the product-mile are not the same mile. Vibe-coding made the first cheap and left the second one where it always was.
Yeah, though I'm not so critical of that part? Obviously there's the whole tokenmaxxing thing, and "when a measure becomes a target is ceases to be a good measure," and all of that, though I think it's sort of directionally right. Or, if you want to do more useful stuff, wastefully moving in the right direction isn't outright wrong, and I suspect we'll get more efficient over time. Though, to your point, perhaps we never move in the right direction at all, and all it is is wasteful.
So in biology, there are 3 major stages when it comes to evolution:
1. Colonization
2. Radiation
3. Sorting
Colonization:
- Who gets in? A small subset of lineages makes it into the newly viable space: survivors of a mass extinction, a few species that cross a barrier, or a single founding population that hits a new island or habitat
- Genetic diversity is low at first (founder effect, bottlenecks).
- Selection favors generalists or “weedy” types that tolerate stress, poor conditions, or narrow resource spectra.
- Traits are often “jack‑of‑all‑trades”: good enough to survive in a patchy, unstable environment, not optimized for anything specific yet.
Radiation:
- What changes? Available niches are wide open, so existing lineages split and diversify rapidly into many new species or ecotypes, each specializing on a different sub‑niche.
- Diversifying selection drives traits to diverge (e.g., beak shape, diet, microhabitat use) so populations don’t stomp on each other’s toes.
- You see adaptive radiation: one ancestor → many descendant species, often with clear morphological or ecological differences (e.g., Darwin’s finches, island plants).
- Time‑wise, this phase can be geologically fast (thousands to a few million years) but still much slower than “boom!” sci‑fi evolution.
Sorting:
- What’s getting sorted?
- Which species persist: competition, predation, mutualisms, and environmental constraints decide who dominates which niche and who gets pushed to the margins or goes extinct.
- Which traits work long‑term: optimization under more stable conditions (e.g., narrower diet specialization, more efficient morphology) replaces the early “generalist” forms.
- Competition intensifies: closely related species or ecotypes compete for similar resources, often leading to competitive exclusion or fine niche partitioning.
- Ecosystems stabilize: carbon cycling, trophic structure, and diversity approach a new equilibrium, but not necessarily a replay of the old ecosystem—new dominant groups and interactions emerge.
From what I observe, now the AI market is either early-mid in the stage of Radiation (you know, if we image AIs as a kind of new creature). My prediction is that, when the AI market mature and it will enter the Sorting stage - probably a few more years if not decades later.
yeah, it definitely seems like we're in the fairly explosive phase, where tons of new things are looking for there niches, and trying to carve out some territory that they can hold on to. And unpredictably so, I'm sure, where things that seem like footholds will probably not last very long, and thing that seem like they are small end up getting very big.
This is strong writing Benn! It has a clear emotional arc and a very modern tension at its core: the gap between continuous building and actual clarity of purpose.
"And when it does, will we be grateful for the places that this delirious moment took us? Or will we wonder where in the world we thought we were going?" I think about this every freaking day.
I think an increasing number of people are asking What’s it all about? What’s the point? Am I happy doing what I do? I think asking the question(s) is valuable and important. The answers are hard. Hell, even finding a place to stand while you ask is hard. This morning in the big Venice fish market I saw 4 seagulls fighting loudly and with great energy over a piece of fish guts. I’m worried that there’s a metaphor there.
there was a whole lot of Discourse about this this week after this whole post (https://x.com/deedydas/status/2055491938464489888), and the seagull analogy perhaps seems better than I wish it were.
We've reached the peak of building for the sake of building.
It was always a problem with the tech industry, and most people working in it, that the emphasis was on the HOW instead of the WHY. Many people have been yelling into the void about "business value" and "solving real problems" for years, and for a while I thought the message might finally start to resonate post-ZIRP era... But no, with LLMs we just completely discarded outcome as a goal and went for maximal output. Or, dear god, for maximal *input* with all the tokenmaxxing talk!
Maybe the eventual turn of the currently exponential hype curve into the usual S-curve will get us back on track. At least I hope so!
It's sort of odd, because pre-AI there was a compulsion to build, but it was a compulsion to build companies. Now, it's more of a compulsion to build *stuff*, which is probably less cynical (there's more genuine fun in that; it certainly could be valuable; etc). But, it still seems rooted in a similar motivation.
I’ve been wondering who would ask the Big Question
I just had to come here to say that ever since reading this piece I think "this is the MOH-ST words" every time Claude writes a 20-page treatise to answer a yes or no question
If you want me to insert some blank pages into my messages to help make this easier digest, I can do that - just say the word and I'll revise it.
So much of it all has been pointless. Unfathomable amounts of wealth wasted for what?
The smartphone at least had immediate obvious implications. It's turned out to be a profoundly terrible invention for society, as it led to the rise of endless social media and all the distortions that has created - but that wasn't obvious immediately, and it has had plenty of actually useful things to at least slightly counterbalance that.
But AI? Nobody has built anything with AI that has truly been Good. We're 3.5 years in and the 'best' we've gotten is using it as fancy autocomplete, or maybe as a tutor to learn stuff - but part of learning is often the struggling to find good info as it helps your brain etch it deeper. All that time, wealth, and talent squandered.
The bubble grows bigger; at some point it must pp. And then we all suffer for their vanity.
I don't think I'm quite so pessimistic about it, that nothing good will come of it, though for the moment, it does sort of seem to be hard to see. Like, it definitely "works," in the sense that it can help people accomplish stuff a lot faster than before, but is that stuff useful? Ehhh? Maybe? My hope, I suppose, is that it goes in the opposite direction of things like phones and social media, where they seemed good at first and then created problems; maybe this one will be somewhat messy at first, and we all eventually figure out how to make it do very useful things that really does make lives better. But I'm not sure vibe coding apps is the thing, just yet.
From all I've seen, AI is...
If you are good at what you do, it's a multiplier. I can draft emails way faster, or auto-customize resumes, that kind of thing (Though so can everyone else, so the arms race gets worse). And for me, that can be helpful - but then I'm not doing the work myself, and I suspect that leads to my own skill development atrophying.
I've seen some limited creative use too - someone who REALLY WORKS at AI art can make good stuff; the Monoverse on Youtube is proof enough of that for me. But you still need taste and a vision. And it's a ton of work.
But if you lack that? You churn out slop. You're just an avatar of the AI, really. And that seems way more common.
Vibe coding apps is definitely not it. What is it requires that grit, the determination to work hard, the caring about the end product...and that lack of care is already so widespread. I'd argue that lack of care is why the economy is rotting; why politicians are so much less responsive; and so much else.
If AI were to help us, the first problem it needs to be tackling is 'How do we make more humans give a shit?' because that's the crippling issue we need to overcome.
And when there's a (apparent?) gold rush, it seems to attract a lot more of that too. That's the most striking thing about that whole Corgi insurance story - not all the stuff about how many hours they work or whatever, but the fact that they're putting in those hours to solve a problem that there's no way in the world they actually care about, other than in the belief that it could produce a big business.
Exactly. There's a ton of 'let's do X, but with AI!!' right now. And I mean, if you really spot a space to innovate? That's fine. But most of these aren't that. They're just attempts to cash in - either on investor money to ride high for a year or two, or even better, a buyout/IPO cause the odds any will be truly viable products are...slimmer. And that may be optimistic.
Video game producers figured this out a long time ago: give people a capacity/"energy" meter, and they won't stop playing when they run out of capacity, but rather they will stay glued to the screen waiting for more, or will even pay real money to get more -- *and they will stay up late at night to make sure that they consume all of their capacity before going to sleep*. And then guess what's the first thing they will do in the morning?
Anything that creates a sense of urgency is good for business (even if it is completely made up).
One more pull.... just one more... I have to, or else I will fall behind!! Time is running out! I must use all the tokens! I can't let any go to waste! There are so many ways to justify it.
AI agents are just the latest game.
But if you were playing tennis, would you have to hit the ball just one more time? Why? What difference would it make?
I've played baseball for a long time, and whenever you're taking batting practice, everyone keeps saying "one more" for about six swings, until you hit one you want to end on. It feels like there's some of that here too, where you don't want to stop until you end at a good stopping point. But everything you add creates a new thing to be done, so that there's constantly one more thing to do before you stop.
The "blurry JPEG of original ambition" is the right framing. Output volume measured in lines-added-to-folder decouples from the actual product question, and the loop where each setback resolves with "I'll rebuild with Opus 4.8" is the same admission at finer time scales. The prompt-mile and the product-mile are not the same mile. Vibe-coding made the first cheap and left the second one where it always was.
Yeah, though I'm not so critical of that part? Obviously there's the whole tokenmaxxing thing, and "when a measure becomes a target is ceases to be a good measure," and all of that, though I think it's sort of directionally right. Or, if you want to do more useful stuff, wastefully moving in the right direction isn't outright wrong, and I suspect we'll get more efficient over time. Though, to your point, perhaps we never move in the right direction at all, and all it is is wasteful.
So in biology, there are 3 major stages when it comes to evolution:
1. Colonization
2. Radiation
3. Sorting
Colonization:
- Who gets in? A small subset of lineages makes it into the newly viable space: survivors of a mass extinction, a few species that cross a barrier, or a single founding population that hits a new island or habitat
- Genetic diversity is low at first (founder effect, bottlenecks).
- Selection favors generalists or “weedy” types that tolerate stress, poor conditions, or narrow resource spectra.
- Traits are often “jack‑of‑all‑trades”: good enough to survive in a patchy, unstable environment, not optimized for anything specific yet.
Radiation:
- What changes? Available niches are wide open, so existing lineages split and diversify rapidly into many new species or ecotypes, each specializing on a different sub‑niche.
- Diversifying selection drives traits to diverge (e.g., beak shape, diet, microhabitat use) so populations don’t stomp on each other’s toes.
- You see adaptive radiation: one ancestor → many descendant species, often with clear morphological or ecological differences (e.g., Darwin’s finches, island plants).
- Time‑wise, this phase can be geologically fast (thousands to a few million years) but still much slower than “boom!” sci‑fi evolution.
Sorting:
- What’s getting sorted?
- Which species persist: competition, predation, mutualisms, and environmental constraints decide who dominates which niche and who gets pushed to the margins or goes extinct.
- Which traits work long‑term: optimization under more stable conditions (e.g., narrower diet specialization, more efficient morphology) replaces the early “generalist” forms.
- Competition intensifies: closely related species or ecotypes compete for similar resources, often leading to competitive exclusion or fine niche partitioning.
- Ecosystems stabilize: carbon cycling, trophic structure, and diversity approach a new equilibrium, but not necessarily a replay of the old ecosystem—new dominant groups and interactions emerge.
From what I observe, now the AI market is either early-mid in the stage of Radiation (you know, if we image AIs as a kind of new creature). My prediction is that, when the AI market mature and it will enter the Sorting stage - probably a few more years if not decades later.
yeah, it definitely seems like we're in the fairly explosive phase, where tons of new things are looking for there niches, and trying to carve out some territory that they can hold on to. And unpredictably so, I'm sure, where things that seem like footholds will probably not last very long, and thing that seem like they are small end up getting very big.
always incredible
This is strong writing Benn! It has a clear emotional arc and a very modern tension at its core: the gap between continuous building and actual clarity of purpose.