I just canceled my @Medium subscription. For a while now, every article on Medium feels like it was written by the exact same person. I can't even finish an article anymore because that feel that every author has an identical voice is so distracting and annoying.
@theAIdreamer@thdxr Hard disagree. The bottleneck was never the code. And that's precisely why, despite almost a year of great harnessess and early access to the best models, the FAANG companies keep moving at the same speed as a year ago, or even slower.
When writing code took longer, our lazy brains had an incentive to pause:
Is this a good idea? Will it create maintenance pain? Do users even need it?
Those questions felt natural because the answers could mean less work.
Now, asking them takes discipline and effort.
I keep hearing the "software engineering productivity" mantra, but either it's not true, or we're wasting all that increased productivity.
I don't see any results. The software I use every day hasn't improved dramatically in the past year. In fact, it's stagnant.
Ken Griffin just asked the question everyone in AI is too scared to answer.
Data center spending in the US this year alone is over $500 billion. Half a trillion dollars. To raise that kind of money, you have to make a promise. And the promise has to be big.
"AI needs to be your savior almost. How else are you going to write $500 billion of checks in a single year?"
He's not saying AI is fraud. He's saying the hype is structurally necessary. You can't fund a buildout at this scale without narrative that matches it.
The real question is what AI actually delivers at the end.
In some areas Griffin says it's going to be profound. Call centers. Software engineering productivity. Those are real, measurable, already happening.
But in white collar work more broadly, he's more skeptical.
A Harvard paper recently coined a term for it: AI Work Slop. Output that looks impressive on the surface. First few sentences read like genuine insight. Then you go deeper and it's all garbage.
Griffin's colleague runs their commodities business. Got handed a report generated by an AI engine. First paragraph, genuinely good. The rest, useless.
The model that can write a compelling opening can't yet think through the substance underneath it.
This is the AI investing tension right now. The infrastructure spend is real. The hype is real. The productivity gains in specific verticals are real. But the blanket assumption that AI transforms every white collar job equally has not been proven yet.
This is where I think coding agents may be failing us. They make implementation feel so cheap that we skip straight to building.
Once we get used to closing 20 issues a day, we might even feel guilty if we stop to think whether those issues are the ones that matter the most.
😱 TINDER: ¿CÓMO SERÍA SI LO HICIESE EL ESTADO?
Es otra de esas ideas que se me ocurren en la ducha tras _días_ frustrado porque un trámite de la Administración electrónica está mal diseñado o no funciona.
¡El Tinder del Ministerio! 😱
🧵👇 Así sería…
So, if gen-ai is giving developers a huge productivity boost, what are developers doing with all that productivity?
I don't see the software I use regularly gaining features or polish faster than before...
Siempre interesante la charla con @el_ryu , esta vez contando un poco de intrahistoria del desarrollo del nuevo juego de @ArumaStudios «Shadows of the Afterland» :-)
Ever noticed how easy it is to quit training, but how hard it is to start again? Your brain naturally leans toward less effort.
What do you think happens when you "quit" thinking hard because a machine can do it for you?