ChatGPT Work vs Claude Cowork: which AI agent should actually do your work?

They're the same idea from two companies: an AI that does the job instead of talking about it. You hand it a folder and a goal, step away, and come back to a finished document, spreadsheet or deck. I've tried both. I didn't love ChatGPT Work. Our company runs on Cowork, and I teach it, so I'm not neutral. Read this knowing that, and check the links, because both tools change monthly.

First, what's an "agent" in plain English?

A chat answers your question. An agent does your task. That's the whole difference. You don't say "write me an email." You say "go through this folder, find the 3 vendors we haven't paid, and draft the follow-ups," and it goes and does it while you do something else. It reads files. It opens apps. It comes back with work, not words. If you've been using AI as a very smart search box, this is the part where it becomes a co-worker.

Both companies shipped their version of this in 2026. OpenAI calls it ChatGPT Work, launched July 9, 2026 (Axios, July 9, 2026). Anthropic calls it Claude Cowork. It launched first, as a desktop tool in January 2026, and reached web and mobile in July (VentureBeat, July 7, 2026). As of September 16, 2026, Anthropic is folding Cowork and regular chat into a single Claude, so you no longer choose up front whether a task is a chat or a job; Claude decides (Fortune, September 16, 2026). The features are the same. The name may fade.

What they have in common

More than the marketing lets on. Both live inside the app you already pay for. Both run in the cloud, and both can reach your actual computer, but only from the desktop app. Both hand back real files: documents, spreadsheets, presentations. Both can connect to your other tools, run tasks on a schedule, and keep separate workspaces for separate projects. Both ask before touching your files, and both let you decide how much they can see. If you're picking between them on features alone, you'll be here all day. The list keeps changing, and it keeps converging.

Where they differ, as documented right now

Everything below is from the vendors' own help pages, checked September 2026. Not my opinion. That comes after.

What it's built for. OpenAI describes Work as "an agent designed for longer, multi-step work and finished deliverables," sitting between Chat and Codex, their coding tool (OpenAI Help Center). Anthropic describes Cowork as Claude for "knowledge work": you "describe an outcome, step away, and come back to finished work" (Claude Help Center). Same promise, nearly the same sentence.

Reaching your computer. Yes, both can. ChatGPT Work runs cloud-only on web and mobile, but on the desktop app it can use "local files and desktop apps with your permission," and OpenAI notes messages "may be stored in the cloud, even when work runs locally" (OpenAI Help Center). Cowork reads and writes your local folders from the desktop app "without manual uploads or downloads," and can browse the web for you through a built-in browser or through Chrome (Claude Help Center).

Running while you're away. ChatGPT Work supports scheduled tasks and event-triggered tasks (a new email, a Slack message, a GitHub change), though event triggers aren't on Free or Go plans (OpenAI Help Center). Cowork's scheduled tasks run "automatically or on demand" in the cloud, and your computer doesn't need to be on (Claude Help Center).

How you organize it. This is the one I care about most, because it's where beginners get lost. Cowork is built around projects, which are separate rooms with their own files, instructions and memory, plus skills that apply to the whole house, and plugins that bundle skills, connectors and helper agents into one package (Claude Help Center). ChatGPT has projects too, and Work sits inside them (OpenAI Help Center). What I haven't found on OpenAI's help pages is an equivalent to skills and plugins, the reusable house rules. That may be a documentation gap or a product gap. UNVERIFIED as of September 2026.

Which plans include it. ChatGPT Work is on Pro, Business and Enterprise, with a limited version on Plus, and it's rolling out gradually (OpenAI Help Center). Cowork is on every paid Claude plan: Pro, Max, Team and Enterprise (Claude Help Center). Neither is on the free tier in any useful form. If you haven't sorted your plan yet, start there.

So which one?

If your team already lives in ChatGPT, try Work first. Switching tools costs more than most people admit, and the agent your people will actually open beats the agent I prefer.

If you're starting fresh, or you're the kind of person who wants to build a real system rather than run one-off tasks, I'd start with Cowork. That's not because it's smarter. It's because the house is easier to build in: one place for your rules, rooms for each client or project, and agents you can hand a job to and trust with the folder. That's how we run our company, with a team of 8+ agents doing the admin I don't want to do. It's also what I teach, so weigh my opinion accordingly.

Either way, the tool is not the decision. The decision is whether you're ready to stop being the workhorse and start handing things off. We are workhorses. The invitation is that you don't have to be.

How to try either one without getting burned

1. Make one folder with nothing scary in it. Give the agent that folder and nothing else.
2. Pick one job you hate that has a clear finish line. "Sort these receipts into a spreadsheet by month." Not "help me with my finances."
3. Read what it hands back like you'd read a new assistant's first draft. Then give it the next job.

Three jobs in, you'll know. Don't worry about the feature list. Worry about whether you'd hand it a fourth.