TL;DR: A new five-platform environment lets you train and evaluate agents that combine GUI operation with coding, and it reveals that even top models look far more capable than they actually are under strict behavioral testing. Title: RecreationWorld: Scalable and Verifiable Environments for Hybrid Computer-Use Agents URL: Key points 🖥️ Covers Ubuntu, macOS, Windows, Android, and Web 🔍 "Recreation" tasks: rebuild a running reference app from scratch, no source code 🧪 RecreationBench: 250 tasks scored via both programmatic and visual assertions 📈 Fine-tuning on 35k recreation trajectories lifts OOD benchmarks by up to 17.9 points 🏆 Top model GPT-6 Astra scores 58.1% overall ⚠️ Yet it passes every programmatic test on only 2.8% of tasks 🔧 Agents flexibly swap Electron apps for native GTK/AppKit without losing fidelity It's a sharp reminder that surface-level recreation and true behavioral fidelity are still worlds apart. #AIAgents# #ComputerUse#

Huge T3 Code release just dropped. Over 380 changes. Quick highlight reel of my favorites: - in-app PR reviews - `npx t3 triage` to debug and fix issues with T3 Code - `npx t3 connect` support on Mac (run 1 command to make a Mac into a remote dev machine) - 93.8%+ performance improvement for rendering long threads - 80%+ reduction to data transferred and stored for threads - In-app terminal performance overhaul - Remote server upgrades are 5x faster and 100x more reliably - Remote projects can now be opened in VS Code via ssh with 1 click - OpenVSX support for custom themes! Use your favorite VS Code themes directly - Codex skill discovery is way more reliable - OpenCode skills actually appear in app now - Claude can discover project-level skills now - Fixed really annoying auto scroll behaviors and edge cases - Claude Code compaction recommendations now surfaced in app - Codex permission requests for MCP/ComputerUse stuff now surfaced in app - Tons of improvements to Windows app - More polish than I can fit into one tweet Sorry this release took so long. I hope it was worth it. Check out the nightly builds if you want to get these features when they're (mostly) done instead of waiting for us to cut a stable release 🫡

Computer-use agent experience = capability × robustness × efficiency. As frontier models climb higher on CUA benchmarks, efficiency becomes increasingly attrative, as highlighted by the recent release of #Jev#. With many intermediate steps in everyday computer-use tasks being “local,” fast System-1 models like #CUA-S1# and #Jev# can make agent execution much more efficient. Cool exploration by @francedot and @ddupont808 at @trycua!

Computer use in @profound Using AI Marketer to analyze the websites of three AI software engineering companies to gather brand styles and page structures. Any task that can be done a computer can now be done automatically in Profound.

Computer use is GA on the Claude API. Stellar production writeup by @davilocatelli13 at Asteroid: agents driving EHRs and payer portals, human steering down from 8.9% of runs to 0% since April.

Computer Use models are rapidly improving, but adoption across enterprises has remained limited. We believe that Browser Agents are an engineering problem, not a research problem. Our founder @pk_iv explains why in his talk at @aiDotEngineer World Fair.

Computer use lets Codex work across your apps without taking over your Mac. @AriX talks with @romainhuet about what changes when agents can click, type, and keep working in the background.