denis

Announcing ARC-AGI-3 The only unsaturated agentic intelligence benchmark in the world Humans score 100%, AI <1% This human-AI gap demonstrates we do not yet have AGI Most benchmarks test what models already know, ARC-AGI-3 tests how they learn

when debating between status-quo and how ai will change (or not) "xyz". IMO the question is always "when" or "how". it has been proven naive to think "human is still better at x" when time and time again ai is able to do it, and many times, better. i understand there are nuances, not saying the contrary, but is useless to debate if we don't consider timeframe. is human better at X today, but for how long? - or - how much will the ecosystem change so that maybe what humans were better at, is not needed anymore as it was. so to have a more productive discussion, timing things help! will ai kill apps as we know them today? I think so. will it happen this year? maybe not. it is already happening? yes if you consider cost to app. does this mean there will be no apps/software at all? IMO there will still be software to be used for a couple of years, maybe even more. but the way we use it is shifting, it is already changing, think IDEs, Claude Cowork, OpenClaw, and will keep evolving and changing, this is just the begining. but it will happen, and fast.

Check out Tech Stack Icons by @keno_reloaded 😍 tech-stack-icons.com

@carlosazaustre convex

Lo que más cambió es que cualquier app nueva va a necesitar conectarse a un LLM. Los desarrolladores que sepan integrar eso, tienen ventaja. ¿Qué le cambiaríais?

Mi stack como AI Fullstack JavaScript Developer en 2026. La diferencia con hace 2 años es que la capa de IA ya no es opcional 🧵👇

Congrats to the @cursor_ai team on the launch of Composer 2! We are proud to see Kimi-k2.5 provide the foundation. Seeing our model integrated effectively through Cursor's continued pretraining & high-compute RL training is the open model ecosystem we love to support. Note: Cursor accesses Kimi-k2.5 via @FireworksAI_HQ ' hosted RL and inference platform as part of an authorized commercial partnership.

just published an npm package ingest data from Garmin, Strava, Whoop, Polar, Suunto, Apple HealthKit, Samsung Health, and Google Health Connect @clipin/convex-wearables" target="_blank" rel="nofollow noopener">npmjs.com/package/@clipi… it's like open-wearables but re-written for typescript/@convex stack

Introducing MiniMax-M2.7, our first model which deeply participated in its own evolution, with an 88% win-rate vs M2.5 - Production-Ready SWE: With SOTA performance in SWE-Pro (56.22%) and Terminal Bench 2 (57.0%), M2.7 reduced intervention-to-recovery time for online incidents to 3-min on certain occasions. - Advanced Agentic Abilities: Trained for Agent Teams and tool search tool, with 97% skill adherence across 40+ complex skills. M2.7 is on par with Sonnet 4.6 in OpenClaw. - Professional Workspace: SOTA in professional knowledge, supports multi-turn, high-fidelity Office file editing. MiniMax Agent: agent.minimax.io API: platform.minimax.io Token Plan: platform.minimax.io/subscribe/toke…

@boshen_c nice!

Taking notes of what Theo says. We'll continue to improve Vite+ 🫡 youtube.com/watch?v=JKbVk6…

@nikitabier last time 4 days, but it's a really small app not sure if that is the reason. using ota updates with expo has been game changer though

iOS developers: How long is App Review taking for everyone these days? It is now taking longer to get our app approved than it is to build the actual features.

High-performance browser control for AI agents. Pinchtab is a lightweight (12MB) Go binary that runs Chrome and exposes a plain HTTP API so any agent or script can navigate web pages, read text efficiently, click/type interactively, and persist sessions. Zero config, framework-agnostic, token-efficient.

depende el caso, ejemplificó algunos casos recoentes - app (nextjs desplegado en vercel) + backend (convex) depslega en su infra - otro ejemplo servicio stand alone (API en golang) cloudbuild.yaml para desplegar en Google cloud + gcloud cli (docker) en cloud run - redis, o rabbit en railway conectado a GitHub gestiona green blue y CD + volúmenes si necesita persistencia - cada vez más hay XaaS databases, queues, etc. planet scale, convex, trigger.dev vercel queues, es como que la infra a pelo es una alternativa más que la única forma. luego dependerá de la escala y aún así estos servicios quieren que escales con ellos (suelen ser competitivos)

Estimados vibecoders con experiencia ¿cómo gestionáis la contenedorización y orquestación en vuestros despliegues a producción? ¿Os abstraéis de la infra/nube donde desplegáis o abrazáis el vendor lock-in como si no hubiera mañana?

@LorenRD @javisantana working on the Spanish version, try estulio.com

@javisantana Try notebook lm with your daughter, it generates games, quizzes, podcasts and videos about any given topic

I see people worried about education and IA and I only see opportunities. - You can ask questions without any shame - Ask for examples - Ask yo explain the topic in a different way - Zoom in/out - Validate hypothesis super fast - Find similar topics - Know where all the information is coming from - Help you to progressively understand topics You can go as deep as you want. I’m pushing my daughter as much as I can to use IA, we ask ChatGPT all the time, sometimes she goes off track and start asking questions about zoombies on the moon but I think even that is positive. Lazy people have always been lazy, with or without IA.

@expo what about android? :D

𝚎𝚡𝚙𝚘-𝚠𝚒𝚍𝚐𝚎𝚝𝚜: docs.expo.dev/versions/lates…

is very interesting how llms estimate tasks they'd complete in minutes

working on a react app? using llms? don't underestimate splitting components across files with a reasonable scope - not because is also better in general for humans - this will allow you to launch agents to work on multiple features at the same time with less overlap

@youyuxi agree

Somehow Codex CLI has a much smoother compaction -> handoff transition than Claude Code

@rohanvarma or related to this, sub-agents/prompts while the main one is working, i can input future changes in sequence "tree-like" ui

@rohanvarma ability to write "draft" prompts and hit enter. with attachments, etc. but not execute them. sometimes when i am working on a thing i want to "stack-up" changes that will come after the other completions finish, but can't stack them. at least to have this done manually

If you have tried the Codex App, but don't use it as a daily driver yet: What needs to improve to make it the primary place you work with coding agents?