11.7 The details on both AI vendors
- Status
- stable
8 min read
- Owner
- Panaversity
- Approved
- Panaversity ·
This page gives the details behind the AI vendor boxes in 11.1 Stop first, then read the record and 11.4 Usage and limits. Anthropic makes Claude. Cowork is where Claude carries out longer tasks for you, and Claude Code is its coding harness. OpenAI makes ChatGPT. ChatGPT Work is an agent that acts across your apps and files, and Codex is OpenAI's coding agent. The details are what each AI vendor's own pages said on 8 October 2026. Products change, so check each AI vendor's current pages before you rely on a line.
Reading the task record
| Anthropic | OpenAI | |
|---|---|---|
| Step list for a task | Progress indicators show each step, and finished outputs arrive in the session.1 | ChatGPT Work: review progress, answer questions, change direction and approve important actions. No separate step log is documented.2 |
| Scheduled runs | Scheduled in the sidebar: upcoming and past runs, each run its own session. Pause, resume, run on demand, edit or delete.3 | Scheduled page: review results, pause or resume. Also reached from Settings > Notifications ("Manage tasks").4 |
| Why a task pauses | In Manual mode a task pauses for approval. In Auto mode it pauses to ask when it blocks an action. There is no list of reasons for scheduled tasks.1 | Needs approval, needs further action, inactive, or its chat was deleted4 |
| A quiet event-triggered task | For Claude Code routines, you can ask /schedule why a routine did nothing, and GitHub events above the hourly limits are dropped.5 | Check for a pause or a pending approval, and confirm the app is still connected4 |
| Standing worker | Claude Tag, in beta on Team and Enterprise, is one shared Claude in a Slack channel. Its Activity page lists every task for Owners.6 | A dot is an always-on agent. Its Activity View, in the desktop app, shows its ongoing and delegated tasks, status and steps. You can redirect it or tell it to stop. Whether an action can be undone depends on the action.7 |
| Changing direction during a task | A message typed during a reply waits. Send now stops the reply and sends yours (added 10 September 2026). In Claude Code cloud sessions, you can go back to an earlier message (added 13 August 2026).8 | Change direction during the task2 |
| Files for scheduled work | Tasks use the connectors and plugins set up for Cowork. Tasks that need files on your own computer run only on that computer.3 | A scheduled task created in a project cannot reach uploaded files or project files.4 |
Usage and limits
| Anthropic | OpenAI | |
|---|---|---|
| Time windows | Fixed-allowance plans (Pro, Max, Team, seat-based Enterprise): a five-hour session limit and weekly limits. Some plans add a separate weekly limit for one model.9 | In Codex, and so in Work: five-hour and weekly windows10 |
| Where to look | Settings > Usage: progress bars for the session and weekly limits, with reset times9 | Usage dashboard. /status in a Codex CLI session.10 |
| What shares the pool | claude.ai, Claude Code and Claude Desktop11 | On plans with shared allowances: Work, Codex, ChatGPT for Excel and workspace agents. Ordinary chat is not on the list.10 |
| At a limit | Wait for the reset, upgrade, or buy usage credits. Credits (Pro and Max) are billed at standard API rates, with an optional monthly spending limit and auto-reload.12 | The active Codex turn can continue, subject to fair use. Then add credits, apply an available reset, upgrade or wait. Support cannot reset limits.10 |
| No fixed limits | Usage-based Enterprise plans, billed by how much is used9 | Enterprise token-based billing: continuing depends on the workspace budget and user limits10 |
| Ways to use less | Content in projects is cached, or stored for reuse, and counts less when reused, until the cache expires.9 Long chats cost more. With code execution on, earlier messages are summarized near the context limit, and those chats use more of the limit.11 | Usage depends on model, task size, reasoning level, speed and tools10 |
| Scheduled task limits | Not documented for Cowork tasks. Claude Code routines allow 100 scheduled runs an hour per account.5 | Free: a one-time task, or a recurring task at most once a day. Paid plans: recurring tasks up to once an hour. Event-triggered tasks: 30 runs an hour, 720 a day.4 |
Check yourself
Question 1 / 8
0 answered
A weekly ChatGPT scheduled task has stopped running. Which of these is one of the reasons OpenAI gives for a paused task?
Sources
- Get started with Claude Cowork, Claude Help Center.↑↑2
- ChatGPT Work and Codex, OpenAI Help Center.↑↑2
- Schedule recurring tasks in Claude Cowork, Claude Help Center.↑↑2
- Scheduled tasks in ChatGPT, OpenAI Help Center.↑↑2↑3↑4↑5
- Automate work with routines, Claude Code Docs.↑↑2
- What is Claude Tag?, Claude Help Center.↑
- Dots privacy, security and safety FAQs, OpenAI Help Center.↑
- Changelog: release notes for Claude Desktop, Claude.ai Documentation.↑
- Usage limit best practices, Claude Help Center.↑↑2↑3↑4
- Using Codex with your ChatGPT plan, OpenAI Help Center.↑↑2↑3↑4↑5↑6
- How do usage and length limits work?, Claude Help Center.↑↑2
- Manage usage credits for paid Claude plans, Claude Help Center.↑
11.6 Make the fix stick, and optimize from feedback
How to move a fix out of a chat so it lasts for everyone, and use feedback and a baseline to improve the work.
Lab 11: diagnose and fix four failed runs
Chapter 11's lab: diagnose four failed runs of Brightline's AP Worker from their task records, fix them, and move the fixes into its setup.