Computer Use

Dustin Hollywood says Stages AI will offer 100 agents, with three agent-controlled editors running at once and coordination across desktop apps. He has set October 22 for the unveiling.

Ozan Sihay says Claude Opus 5.5 edited the opening 80 seconds of his YouTube video in Premiere, including animations. He reports using about 30% of his session quota at high effort.

Anthropic released Haiku 5.5 with a reported tenfold price reduction for requests below 100,000 tokens. Aakash Gupta cites $0.10 input and $0.50 output per million tokens and a 72.4% OSWorld score; Viktor Oddy demonstrates the model on website design.

Higgsfield demonstrated its ChatGPT plugin using computer controls to turn source material into creative deliverables. Demos included YouTube Shorts, music-release campaigns and client work.

OpenAI’s Dots can keep working across connected apps while users are away. Higgsfield says three Dots split an After Effects project into scene, animation and styling tasks, finishing in 15 minutes for $30.

Google introduced Gemini 3.8 Live and Live Extended Thinking in Gemini Live and its API. The models add near-real-time visual understanding, support for 97 languages, and background tool use.

Higgsfield demos show GPT-6 Astra drawing in Photoshop and Procreate, recreating motion work in After Effects, editing in Premiere, and making an editable Figma poster. Other creators report using Astra-generated Adobe scripts to avoid edits.

Creators report Astra writes scripts and carries out edits in Adobe apps. Demonstrations show it operating Photoshop, Illustrator, After Effects, and Krita through desktop or MCP workflows.

OpenClaw says users have moved to its web interface and native apps after a major rewrite. The project acknowledges one report that its migration tool missed legacy exec rules.

DeepSeek's experimental V4 Flash Vision model is reportedly live on its official API. In a screenshot-only form-filling test, it completed and checked fields in 5 minutes 48 seconds but misplaced check marks and drifted in longer boxes.

Users report deploying Grok Bot for churn recovery, support inboxes, refunds, expense reduction, and sales prospecting. One report says it found $175 a month in subscriptions to cancel after obtaining approval.

A roundup of user examples says Grok Bot can build games, manage repositories through recruited agents, SSH into a Mac, buy domains, and deploy projects. The reports describe computer-use workflows extending from code generation to deployment.

Tutorials and user reports show Grok Bot handling web research, inbox cleanup, travel planning, and game installation through a persistent cloud computer. The trade-off is a focused, responsive experience against lower flexibility, a reported $200 monthly starting price, and lag during gameplay.

Developers proposed agent-specific browser auth for Codex, Devin, Claude, and Cursor so agents can use selected passkeys, TOTPs, and passwords without exposing full vaults. Related posts warned that Claude Connectors may broaden tool access.

Allie Miller described using Codex Voice Mode for trackers, email, reminders, scheduling, and document work. Min Choi’s ElevenLabs setup shows a browser agent that listens, streams replies, and handles interruptions.

Peter Yang shared Jason Liu’s workflow for mining past sessions into skills, adding a /write-like-me skill, pinning durable threads, importing browser sign-ins, and bundling related skills as plugins. One example edited a launch video from Slack feedback.

Users reported the former Codex Superapp now appears as ChatGPT Work, with Codex inside the ChatGPT desktop flow. Early posts flagged broken sites support and confusing sidebar states.

OpenClaw's maintainers say the community-built iOS and Android apps are live, with secure pairing and push notifications. The rollout also exposed staffing, funding and plugin-support limits, so the project is still looking for contributors.

Steipete showed Codex running inside Crabbox, opening accounts and completing live web signups while building and end-to-end testing the same project. The setup extends coding agents into real browser actions, but it can also trigger payment and verification messages during automated runs.

OpenClaw maintainers said Microsoft showed the OpenClaw gateway at Build and tied it to Windows Execution Containers for native sandboxing. The observability and verifiable-workspace features push the project closer to enterprise computer-use deployments on Windows.

Google DeepMind showed an experimental pointer that lets Gemini act directly on screen elements with motion, speech, and shorthand commands. The demos move assistance from chat into live workspace control, but the feature was presented as an experiment rather than a shipped product.

UI-TARS resurfaced as an open-source desktop-control stack while Opendesk described using accessibility APIs and marked elements instead of raw pixel guesses. The approach makes computer-use workflows more repeatable, but it still depends on human-oriented interfaces.

Levelsio added a Cursor-style sidebar to Photo AI that uses Grok 4 to take photos, run packs, and remix uploads by operating the web app’s own controls. The demo stored chat history in localStorage and used on-screen context to translate requests into UI actions.

Codex App Server added a Fedora RPM package for Linux installs as users pushed Codex into browser control, 3D-print setup, and rapid game prototypes. Watch for more repeatable desktop workflows as Codex moves beyond chat-only experiments.

OpenAI updated Codex with Mac app control, background computer use, image tools, ongoing tasks, and 90+ plugins, while Remotion added a one-click skill. Agents can now work inside desktop creative apps and stacks without blocking the visible cursor.