ChatGPT Work: A Cloud Agent Built on the Codex Harness That Books Travel
Melvin Vivas · X video post · 2026-09-04 · 1:00 · 385 views · Open on X
Topics: AI Agents, Tool Use & MCP, AI Dev Tools & Productivity, Industry Trends & Job Market · Level: beginner
Summary
Melvin Vivas reshares an OpenAI demo of ChatGPT Work, a cloud agent that runs on the Codex harness. In the demo, the agent goes to travel websites, signs in with the user's accounts, finds a flight paid for with miles and a beach apartment within budget, and books both after the user approves. It shows how a coding-agent harness can run general browser tasks for users who aren't developers.
Key points
- ChatGPT Work runs in the cloud on the Codex agent harness, so the same agent loop used for coding also handles general web tasks.
- Example request: find a flight from New York to Punta Cana on October 17 and check whether the user has enough miles to cover it.
- The agent goes to the website itself. When it needs a login, the user types the credentials, then the agent keeps searching using the user's miles and preferences.
- If a site session already exists, the agent skips the login and goes straight to the task. In the demo, this was an apartment near the beach that was available on the dates and within budget.
- Human in the loop: the agent shows the flight and the stay for review, and books only after the user approves.
- The agent returns booking confirmations at the end. The demo presents this as planning and booking in one place, even while away from a desk.
- Design point for agent builders: handling login, reusing sessions and asking for approval before money is spent all matter for agents that take real actions.
Resources mentioned
- ChatGPT Work · tool · chatgpt.com · check price
ChatGPT workspace on web and mobile for creating docs, decks, sites and spreadsheets and handling complex tasks.
Also in: ChatGPT Voice Update: Plugins, Model Switching and ChatGPT Work (Melvin Vivas on X · notes), Hermes (Grok 4.6) vs ChatGPT Work on a text-to-PDF task (Melvin Vivas on X · notes), ChatGPT Voice on Desktop: Directing Codex and Agents by Voice (Melvin Vivas on X · notes) - OpenAI Codex · tool · openai.com · paid · recommended by both Bashiri Smith & Melvin Vivas
OpenAI's coding agent. In the diagram it writes code, fixes review findings and drives the build loop. The creator also used it to make this video.
Also in: Developers moving from ChatGPT/Codex to Claude (Melvin Vivas on X · notes), Creator Opinion: Codex Is Still the Best at Computer Use (Melvin Vivas on X · notes), OpenAI Codex Ambassador applications are open (Melvin Vivas on X · notes), GPT-Live-1.1: Multilingual Voice Update for ChatGPT and Codex (Melvin Vivas on X · notes) and 264 more - ChatGPT · tool · chatgpt.com · free · recommended by both Bashiri Smith & Melvin Vivas
OpenAI's AI assistant (web, desktop and mobile apps), used for chat, coding, writing and image generation.
Also in: Keep a Change Log When Multiple AI Agents Edit One Notion Doc (Melvin Vivas on X · notes), Multi-Agent Planning in Notion: Grok Bot, ChatGPT and Claude (Melvin Vivas on X · notes), Hark Pro Launch: A Proactive Personal AI Agent with Cloud Computer Use (Melvin Vivas on X · notes), Why Grok Bot Could Be the Next "ChatGPT Moment" for Personal AI Assistants (Melvin Vivas on X · notes) and 35 more
Try this
- Try ChatGPT Work (Codex in the cloud) on a real task that takes several steps on the web.
- Give the agent a task with clear limits (dates, budget, preferences) and check its results before letting it book or pay.
- Build a travel-booking agent that searches flights and stays, checks loyalty miles, and asks for human approval before it books.
More in AI Agents, Tool Use & MCP
- Coworker: An Agent App Inspired by Grok Bot, Built on Pi
- Coworker: Open-Source Desktop App for Local-Model AI Agents
- Docs7: Serving Agent-Readable Documentation to Coding Agents via MCP
- Share Pi Agent Traces on Hugging Face with pi-share-hf
- Wish: Automatic Model Selection for Codex Subagents
- JSONL Viewer v1.3.0: View Codex, Claude Code, Pi and Hermes Agent Traces