How to Use ChatGPT Work: Access, Plans and What It's Actually For

Getting to it takes one click:

If you don't see it, check your plan.

Who can use it

The desktop app is the interesting row: Chat, Work and Codex are all there on the free tier. If you want to try an agentic mode without paying anything, that's the door.

Retrieved 20 September 2026.

What it's for

Work takes a brief, not a question. You describe an outcome, it goes away for minutes or hours, and it returns a finished artifact — a spreadsheet, a slide deck, a report, a small working web app. It reaches your tools through a plugin directory (Slack, Gmail, Drive, Salesforce and dozens more) and it can run on a schedule.

The mental model that makes it click: you're not chatting, you're assigning. Prompts that work well in Chat often underperform in Work, because Chat rewards a good question and Work rewards a good spec.

A brief that works looks more like:

Pull the last four quarters from the attached exports, build a summary sheet with quarter-over-quarter deltas by region, flag anything moving more than 15%, and give me a five-slide deck with the three biggest movers.

…and less like "can you analyse this data?"

Three modes, easy to confuse

• Chat — conversation. Ask, get an answer.
• Work — multi-step knowledge work, returns an artifact.
• Codex — the developer coding mode.

The older application is now ChatGPT Classic. If someone tells you to "use agent mode," they most likely mean Work.

How the metering works

Work isn't a separate purchase. It draws on a credit allowance shared with Codex.

Normal use doesn't hit it. A full day of heavy building can. When you run out, you either wait for the allowance to refill or buy extra credits.

Worth internalising: an agentic run is dozens of model calls and tool invocations, so it consumes far more than the equivalent time spent chatting. Budget accordingly if you're planning to lean on it.

Working across devices

Work on web and mobile runs in the cloud, and those chats sync. A Work chat started on web or mobile appears in the desktop app, and a cloud Work chat started on desktop continues on web or mobile.

So the practical pattern is: kick off a long job, close the laptop, check on it from your phone.

Getting more out of it

Write specs, not questions. Name the inputs, the output format, and the decision rules. Work executes plans well; it doesn't guess at intent well.

Say what "done" looks like. "A five-slide deck" beats "a summary." The product is built around returning a finished thing, so tell it which thing.

Connect the plugins first. Most disappointing first runs are Work being asked to reach something it has no connector for.

Use the schedule. Recurring briefs — a Monday pipeline summary, a monthly reconciliation — are where the horizon actually pays off, and most people never turn it on.

Don't use it for quick questions. That's Chat, it's faster, and it doesn't burn credits.

The one limitation to know

Work lives in ChatGPT. Every route to it — web, mobile, desktop — means opening OpenAI's app. There's no way to reach it from iMessage, WhatsApp, or your team's Slack channel.

For most people that's fine. If it isn't, that's the specific gap worth shopping for, and it's what we build at Agentastic.

---

Comparing against the alternative: Claude Cowork vs ChatGPT Work. Sources: ChatGPT Work and Codex, openai.com/chatgpt-work. Retrieved 20 September 2026.