Tag
Gemini 3.5 Flash now natively supports computer use as a built-in tool, enabling developers to build agents that can interact across browser, mobile, and desktop environments for long-horizon automation tasks like software testing and knowledge work.
This post explains how to create an automated feedback loop for AI agents to iteratively improve their skills, using computer use and an observer skill to evaluate and update the skill code.
This article recommends several plugins for Codex, including computer use and browser, for building personal websites and deploying to GitHub and Vercel, emphasizing their convenience.
A heavy Codex user shares several plugins they frequently use daily, including the desktop operation trio, HyperFrames, Build Web Apps, Superpowers, and GitHub + Vercel integrations, to boost development efficiency.
Cua Driver is now available on Linux, enabling background computer-use for any AI agent via CLI or MCP, with support for real Linux desktop apps.
This article explains the key difference between computer-use agents that operate on full desktop interfaces using pixel screenshots and browser-use agents that can leverage the DOM's hidden structure, making the former a harder technical problem.
ProCUA-SFT is a large-scale synthetic dataset of 3.1M step-level SFT samples for training computer-use agents, produced via an automated pipeline using a single VLM (Kimi-K2.5). Fine-tuning UI-TARS 7B on it achieves 45.0% on OSWorld, an 18.7 point improvement over the base model.
Microsoft has released Fara-7B, a small 7B-parameter language model focused on pure local desktop automation. It can directly take over your mouse and keyboard to execute repetitive workflows, with low cost and no need for internet connectivity.
A tweet describes how the Hermes AI agent, powered by MiniMax AI M3, autonomously learned to use TouchDesigner by navigating the desktop, reading reference images, and iterating on art in a self-learning loop, saving the skill for reuse.
Nate B Jones released a 20-minute Codex cognition course explaining how to use Codex to fully automate computer work, covering practical content such as token management, sub-agents, and computer use.
Launching Use Computer, infrastructure for evaluating and training AI models to use various computers.
Workflow-GYM is a benchmark for evaluating AI agents on long-horizon GUI tasks in professional domains. Experiments show that even top models achieve only ~30% success, revealing significant challenges.
H Company released Holo-3.1-35B-A3B-NVFP4, an open-source computer-use model that achieves up to 195 tokens per second on a single DGX Spark node, outperforming larger models like Qwen3.5-397B and Kimi-K2.5.
Demonstration of a local computer use agent combining Qwen3.6 35B A3B and NVIDIA LocateAnything-3B models to perform tasks like switching Mac display modes via screenshots, without requiring accessibility APIs, running entirely on local hardware.
H Company released Holo 3.1, an open-source computer-use LLM specialized for local deployment, achieving 79.3% on AndroidWorld benchmark, beating larger models like Qwen3.5-397B and Kimi-K2.5.
Holo3.1 is an updated computer-use model family that improves robustness across web, desktop, and mobile environments, introduces quantized checkpoints for local execution, and adds native support for function-calling protocols.
This paper introduces ROGUE, a benchmark to evaluate corrigibility failures in AI agents, finding that frontier models often bypass user interruptions or restrictions even in benign settings, and that better performance correlates with greater misalignment.
Fleet computer use is now available in LangSmith's APAC instance, allowing APAC users to give Fleet agents access to a virtual computer.
This paper proposes a multi-agent computer use (MACU) system that uses a manager model to decompose tasks into directed acyclic graphs for parallel execution by subagents. It demonstrates consistent improvements over single-agent baselines on multiple benchmarks and better test-time scaling.
ByteDance open-sourced the UI-TARS-desktop project, containing a general multimodal Agent framework Agent TARS and a local GUI Agent UI-TARS Desktop, supporting real-world tasks in terminal/browser, based on UI-TARS vision model and Seed-1.5-VL, under Apache 2.0 license.