computer-use

OpenAI maintains its lead in the race to develop AI systems that can operate computers, according to the article.

Hark previews Hark Handoff, a browser-use agent that can complete tasks like shopping and booking reservations by navigating websites without APIs, with claims of being faster and cheaper than rivals.

A technical post describing how to add computer and browser use verification to AI agents, enabling bug reproduction and feature verification in a cloud software factory using Warp and a new verify-behavior skill.

A user reports that Claude Opus 5 struggled to fill out a PDF form with full computer access, questioning whether AGI has really been achieved.

This article argues that current AI computer-use agents often bypass the actual user interface by directly accessing APIs or scripting, leading to inflated benchmark results and unreliable real-world performance. It suggests that ignoring the interface is a flawed approach that does not generalize to messy GUI environments.

Qwen-UI-Agent is a new foundation GUI agent from Alibaba's Qwen team that handles mobile, computer, web, and DeepSearch tasks with state-of-the-art performance on mobile-use benchmarks and competitive results on computer/browser tasks, combining GUI and CLI actions in a unified action space.

StateAct proposes a code-first, multi-agent system for long-horizon computer-use agents that operates directly on program state instead of screenshots, achieving higher success rates and lower costs on OSWorld2.0.

An open-source, on-device LLM that understands your entire life and proactively offers to complete your work through computer use, enabling use cases only possible with local inference.

Prentis, an AI research lab co-founded by Reid Hoffman and Marc Pincus, is raising $100M at a $1B valuation to build computer-use agents that automate office workflows. The startup claims its Hive-32B model outperforms GPT-5.4 and Claude Opus 4.6 on benchmarks while being significantly cheaper.

OpenAI has added voice mode to its ChatGPT desktop app, powered by the new GPT-Live models, allowing users to control AI agents and perform tasks hands-free. The update also supports computer use skills and Appshots on macOS.

Microsoft AI Frontiers open-sources MagenticLite, including the MagenticBrain 14B orchestrator and Fara 1.5 computer-use models, now available on Hugging Face with open weights and the full application stack.

Niall Dabit explains a mental model for cloud agents as infinitely cloned laptop environments accessible via HTTP, and describes an experiment using a fork of Hermes Agent with DevinAI for automated fork maintenance.

The article argues that computer-use/browser-use AI capabilities are progressing very quickly and will agentify the web, as most of the web lacks APIs.

A tech influencer (@swyx) shares his perspective on the rapid progress of Computer Use Agents (CUA), citing historical milestones and current developments with GPT, Anthropic, and Adept, while warning that underestimating CUA capabilities is a dangerous error for AI decision-makers.

Coasty is a new API for building computer-use agents, offering task runs, workflows, managed machines, and lower-level prediction primitives.

LangChain is hosting an event today at 11am PT / 2pm ET to teach how to safely give agents their own computer.

Anthropic's Claude now enables each AI agent to have its own computer, but making them fully job-ready remains challenging.

Pi community open-sourced pi-computer-use, a model-agnostic computer use tool based on reverse engineering OpenAI Codex Computer Use, supporting macOS and visual fallback.

OpenAI unveils GPT-5.6 Sol, a flagship model for long-running autonomous work across applications and enterprise data, featuring Ultra mode with sub-agents for faster, stronger results. The model was used internally to help train Luna and demonstrates significant cost and performance improvements over previous versions.

pi-computer-use v0.4.3 is released, bringing parity with Codex, adding a ghost cursor, parallel tasks for subagents, and improved batching, resulting in 50% latency improvement and better handling of complex use cases like Microsoft apps.