Yago Mendoza

Agents are out there, renting GPUs and working on PyTorch right now.

YC on how to build a company with AI from the ground up:

@vercel_support @vercel Pro customer here. Locked out of Dashboard for 15h after Pro upgrade. Security Checkpoint sfo1:: blocking all browsers/devices. Can't even access help.vercel.com. Ticket number or DM for support?

@jackfriks Working hard ❌ Hardly working ✅

for the first time in human history you can do 1 week of work in 1 hour with claude and 99% of people use this chance to work more instead of less

it even threw a little ladybug 🐞 in the corner as the toggle. unrequested

the move is having your AI spin up a draggable override debug panel that patches your app's state bus directly, lies to other components and lets you hot-swap sync states to bypass the real signal. zero test infra, no mock yaml configs, no devtools tabs.

Not whether they memorized Dijkstra's algorithm. Whether they can architect, prompt, debug, iterate, and ship. Today. Right now. The skill that matters this year, next year, and every year after.

What if instead you just gave candidates multiple screens, internet access, their own subscriptions, Claude, Cursor, whatever they're paying for and said: "Here's a spec. You have 4 hours. Ship as much of this app as you can"

You're spending engineering hours building proctoring software to stop candidates from using AI? You've successfully filtered for people who can reverse a linked list from memory in 2026. Like hiring pilots based on who can fold the best paper airplane. Kudos.

@_guillecasaus GGUF?

Acaban de conectar células cerebrales humanas a un modelo de lenguaje. Cultivaron 200.000 neuronas en laboratorio, las integraron en un chip y ahora están enviando impulsos eléctricos reales para decidir cada palabra que genera la IA.

@BrosnanHoban @iruletheworldmo I mean, from an actual meaningful societal progress POV, it is

@iruletheworldmo already mainlining AI like it's oxygen 💨

another incredible article on cowork. i guarantee you’re not getting maximum value from ai and after reading this you will be. read it right now sir. don’t you dare bookmark and forget

@MLDataTorturer @iruletheworldmo at least we get to touch some grass, glass half full

@iruletheworldmo I'm not getting the value because it's down constantly.

@GusevV1987 @signulll exactly this. the winning move is making the agents disappear

@signulll The winning products will abstract “agents” behind simple user experiences, just like apps did before them.

the term "agent" will likely never become "app".. it will stay in developer jargon & never reach civilians. which is how it should be. lemme explain why. people understand "app" because apple built a distribution surface that made the concept self evident, then spent billions hammering "there's an app for that" until the word fused with the thing. no one is doing that for agent, & no one needs to. this is because ai isn't a new category of object. also it's not a noun waiting to be named. apps were discrete, bounded, downloadable, & pointable at. ai is ambient. it dissolves into the infrastructure of everything else. you don't need a word for something that becomes the medium rather than the message. imho the right analogy is electricity. nobody uses an electricity. you just flip the switch & the lights come on. ai will be the same. the interface disappears, the verb changes, & the underlying orchestration becomes invisible. the people trying to make agent happen are making the same mistake as people who tried to make horseless carriage stick. they're describing the new thing in terms of the old thing. the new thing will eventually just get called whatever it does.

@geladaris_georg @signulll and that device will be what we call a product & the agents behind, invisible plumbing. @signulll is just naming what the surface will be called before we have the word for it yet

@geladaris_georg @signulll or a device. a single surface. behind that surface, sure, maybe a thousand agents are orchestrating, burning through tokens like electricity, but the product is the illusion that there's just one thing talking to you. one device. pure AI

world question: in a knowledge graph with both tree hierarchy AND cross-links, what's a good heuristic for "relevant neighborhood"?

If this is you, reach out. DMs open. RTs appreciated.

First step: publish an evaluation standard before April's EU consultation. Bringing 3-4 colleagues for co-authorship is a huge plus.

More vulnerable people use AI for mental health without clinical evaluation. We're building Europe's first company to fix this.