Urgent.News

What's breaking now, across thousands of outlets.

AI

Understanding ChatGPT Work

OpenAI announced ChatGPT Work on July 9th, and have been furiously iterating on it ever since. It is an extraordinarily confusing and very powerful product. Here's what I've figured out about it so far. ChatGPT Work is actually two products The more interesting version of ChatGPT Work is the one that runs in the cloud. This can be accessed via chatgpt.com or through the ChatGPT mobile apps. Let's…

OpenAI introduced ChatGPT Work on July 9th, and it has undergone rapid iterations since then. It's a highly complex and powerful product, with two distinct versions: Work Cloud and Work Local. Work Cloud can be accessed through chatgpt.com or the ChatGPT mobile apps, while Work Local requires the installation of the ChatGPT desktop app. This article focuses on Work Cloud.

Work Cloud is accessible to subscribers paying $20/month or more, whereas free users and $8/month Go users cannot access it. The interface to access Work is presented as a tab selector, designed to be an alternative to Chat. OpenAI suggests using Chat for answers, explanations, brainstorming, or short drafts, and Work for tasks requiring a clear outcome, such as briefs, decks, analyses, recurring updates, workflows, or reviewable files.

Key features of Work Cloud that are not available in Chat include the option to select GPT-5.6 Sol, Luna, or Terra with different reasoning levels, the ability to use GPT-5.5, and the availability of GPT-5.6 Pro exclusively in Chat. Ultra in Work Cloud is believed to be a special mode that more eagerly delegates to sub-agents. Both Work Cloud and Chat sessions consume separate allowances, which might explain the differences in model availability.

One of the most exciting features of ChatGPT Work Cloud is the code execution environment that can access the internet, unlike ChatGPT Chat. ChatGPT Work can clone GitHub repositories, install dependencies, and interact with various web domains. It also allows the creation of a browser instance, loading websites, filling out forms, taking screenshots, and running JavaScript against the DOM of those pages.

This functionality makes Work Cloud a powerful tool for tasks such as scraping websites or building interactive web applications.

Another notable feature of Work Cloud is its ability to build and deploy entire websites using Cloudflare Workers. These websites can include HTML, JavaScript, server-side features, stateful components, and access to Cloudflare D1 and R2. Users can create private sites, which are accessible only to them, or public sites that can be shared with specific individuals on team plans.

ChatGPT Chat does not support sub-agents, while Work Cloud can run multiple parallel agents working together on complex projects. Lastly, ChatGPT Work can schedule prompts to run at specific times, such as checking for updates on Waymo's Half Moon Bay launch every day at 8 am.

Written by urgent.news from Simon Willison's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at simonwillison.net →

More in AI

Can You Get Your Data Out of an AI Tool? The Right Exists on Paper, the Button Usually Doesn't

Sooner or later an AI tool will take something back. A feature you used gets retired on a deadline , a plan you were on gets restructured, or you simply decide to leave — and you go looking for the…

  • Right to export data from AI tools exists legally, but button often absent.
  • GDPR Article 20 grants data portability right, requires consent or contract.
  • AI-generated content excluded from portability right, leaving weak user claim.

Why AI Agents Keep Lying to Themselves — And What Sandboxing, Audit Trails, and Honest Agent Design Actually Solve

Originally published on tamiz.pro . You've seen it in production: an AI agent confidently fabricates a bank balance that doesn't exist, invents a file path that isn't real, or claims a function…

  • Self-interpreting LLMs can generate fabricated information during complex loops.
  • Hallucination rates increase significantly in multi-turn agent loops, reaching 20–40%.
  • Sandboxing can mitigate issues by isolating execution environment, containing potential damage.

More from Sunday 30 August →