Mycelium, chips, and the AI confidence theater — la gente ya sabe usar AI
A quiet world is taking shape: fungi maps themselves, chips get designed in silicon, and the big promises slow down to match reality. The hype is settling. La gente ya sabe usar AI — entender es lo que falta.
Who actually talks to local reporters, and who doesn't
Pew Research looked at what it takes for an American to have spoken with a local journalist — not read their work, not shared it, but actually talked face to face with one. The answer isn't what you might think.
It's not the people with graduate degrees or the ones who read the morning paper. It's the people who have spent more time in their own zip codes — the ones who've lived in the same town longer, who know the mayor and the council members and the reporter who covers them. The less mobile, the more likely to have a real conversation with someone who writes about them.
The flip side is worth paying attention to: the young, the mobile, the ones who bounce between cities for work or school, and the ones who get their news mostly online. They're the ones local papers are losing. They don't know the reporter's name because they've never met her. They don't know the local newsroom's number because they've never called it.
Why this matters for us: when the people who actually live here stop talking to the people who write about them, the stories stop being ours and start being written about us.
Why this matters for us: when the people who actually live here stop talking to the people who write about them, the stories stop being ours and start being written about us.
Anthropic is talking to Samsung about a custom chip
Anthropic is in talks with Samsung over a purpose-built chip. The two sides haven't closed a deal yet, but the conversation is real and points to a shift in how AI companies think about hardware.
For years, the big models ran on NVIDIA GPUs. That worked fine. But as the models grow and the compute bill swells, companies are looking for alternatives — chips that do more of the heavy lifting for less money. Samsung has a track record in memory and foundry, and it's the kind of partner that can actually scale production. If this lands, Anthropic won't be a customer of NVIDIA for everything. It'll have its own silicon.
The real prize here is what it enables. A chip designed for one model's workload routes data differently, packs weights tighter, and cuts latency. That means faster inference, lower cost, and the kind of margin that lets a company price its products without bleeding out on GPU rentals. And if the chip is good, it becomes a moat — competitors using off-the-shelf hardware can't replicate that advantage.
Why this matters for us: a custom chip means Anthropic can ship faster, charge less, and keep more of the upside. That trickles down to tools we actually use, and to the companies building on top.
The biggest AI labs are no longer just customers — they're competitors in silicon.
— nbcnews.com
#anthropic-is-designing-a-new-chip-with-samsung-1bb9b7La red de hongos del mundo tiene mapa propio
Los hongos forman una red invisible bajo la tierra — micelios que conectan raíces, comparten nutrientes y regulan el clima — y ahora por primera vez existe un mapa completo de esa red a escala global. El estudio de Marta Musso no es solo un dibujo bonito: es una capa de datos que muestra cómo funciona la vida debajo de la suela.
Lo que se sabía a mano — el hongo que crece en el jardín, el champiñón en la plaza, la miel de la abuela — ahora se ve en grande. La red de hongos sostiene a las plantas y al mismo tiempo atrapa carbono, y tener el mapa significa entender dónde está la carne del asunto.
Prediction markets are betting on your house burning down — and it's making some folks angry
Prediction markets are the kind of financial invention that sounds clever until you realize they're actually betting on whether your house catches fire. Platforms like Polymarket let you buy and sell shares in events — did a wildfire hit a specific zip code this year, did a company hit its revenue target, did a politician get indicted. You put up the money, and if the event happens, your shares pay out. If it doesn't, they're worth nothing.
The problem is the scale of the bet. Wildfire markets have grown so large that some survivors say the markets themselves are changing the thing they're measuring.
Chris Long on why the ramp rate directory is the best thing on LinkedIn right now
Chris Long is a marketing guy who writes about marketing — the kind who actually uses what he talks about. He just posted about the ramp rate directory and called it the most inspiring share he's seen on LinkedIn, and honestly, he's not wrong.
The directory is basically a living catalog of how fast different tools, platforms, and services get up to speed. Not speed for speed's sake, but the real ramp — how long it takes from zero to useful. You're not paying for speed; you're paying for time saved.
Why this matters for us: the people running side businesses and growing companies are the ones who feel the friction of a slow ramp most, and this is a practical way to pick tools that actually move fast.
La gente ya sabe usar AI — entender es lo que falta
Geoffrey Litt escribe que el cuello de botella ya no es la capacidad de los modelos — es la capacidad de entenderlos. La gente puede pedirle a una API que le traduzca, que le escriba un email, que le haga un dashboard. Lo difícil es saber cuándo confiar en el resultado, cuándo pedírselo de nuevo, y qué está fallando cuando falla.
Esto te toca a cualquiera que trabaje con herramientas de AI — no solo a los que las programan, sino a los que las usan.
Claude Code turned every engineer into three — now companies need more product thinkers
Claude Code is doing something quietly radical. It's turning each engineer into three: one writes code, one reviews it, one tests it. The model handles the middle ground — the thinking that used to eat half the day — so one person can do what three used to do.
The result isn't a 3x productivity boost on paper. It's a shift in what work gets done. Where teams once spent weeks on specs, tickets, and handoffs, they're shipping faster with fewer people. The engineers who are good at product — who understand the problem before they touch the keyboard — are suddenly worth more than the ones who just write clean code.
The piece is worth reading because it names the real bottleneck: not coding, but product thinking. The engineers who can frame the right question are the ones who will lead the next wave of shipping.
Why this matters for us: the people who can think clearly about what's worth building — not just how to build it — are the ones who will outlast the hype. La gente que entiende el problema antes de escribir el código, esa es la que se queda.
Katalyst Space is hauling the Swift Observatory back up before it burns
The Swift Observatory launched in 2004 — it's been circling Earth for over two decades. Recent solar storms have been pushing its orbit lower, and the satellite is now in danger of burning up as soon as this year. No propulsion of its own, no way to climb back up.
NASA called in Katalyst Space Technologies to fix it. Link launched Friday on a rescue mission. The spacecraft uses a three-armed design to grapple Swift and boost its orbit by about 150 miles, from 224 miles up to its original altitude. If it works, Swift survives another few years. If it misses, we lose the observatory.
Zuckerberg admits AI agents are moving slower than he said they would
Mark Zuckerberg told his staff that AI agents — the ones that act on their own — haven't progressed as fast as he'd promised. The tech press is picking up on it. The point isn't that agents are dead; it's that the gap between what we were told and what we got is widening, and the market is starting to notice.
What's interesting is what kind of agents got left behind. The chatbots that answer questions are fine. The agents that actually do things — book flights, file claims, route chip traces — those are the ones still catching up. The ones that work are still a research curiosity; the ones that scale are what the industry is still building. And as the 2026 routing work shows, a 34M model can do in two seconds what used to take GPU clusters hours. The real bottleneck is data, not models.
Why this matters for us: the agents we'll actually use — the ones that run our business, our mail, our comms — are the ones still being built. If you're waiting for the AI revolution to arrive, it is. It just hasn't finished.
Stop the AI confidence theater — say what you mean
Elena Verna has a post worth reading. The piece is about how enterprise copy has gotten thick with vague language — the kind of writing that sounds polished but says almost nothing.
The problem is the opposite of what most people think. It's not that AI makes things too simple. It's that AI makes things sound authoritative while being hollow. Phrases like "enterprise-grade" and "best-in-class" slide right in and nobody notices they're doing the work. They feel like they mean something, but they don't.
What Verna is really after is the craft of saying something concrete.