ChatGPT Work
The agent side of ChatGPT, built for longer, multi-step jobs and finished files.
- Free option: Yes, desktop only
- Open source: No
- Worldwide (where ChatGPT works)
Inbox, coworkers, daily priorities.
In the real examples we've collected, people mostly use ChatGPT Work (2 examples here, paid from $0, where: worldwide (where chatgpt works)), Claude Cowork (1 example here, paid from $20/mo, where: where claude is available) and Make (1 example here, paid from $16/mo, where: most countries) for email & work. Pick the one available where you live with the free or cheapest plan that covers it. We haven't tested agents on this job ourselves yet.
The agent side of ChatGPT, built for longer, multi-step jobs and finished files.
Claude's mode for doing tasks, not just answering. Runs in the cloud and can use your files and browser through the desktop app.
Connect your apps step by step on a visual canvas and add AI where a decision is needed. Hosted only, with a free plan that never expires.
The poster had their ChatGPT Work and Claude Cowork agents go through the SF and LA Tech Week events and register them for the ones that fit, based on what the agents already knew about them. It was a quick, happy post, so it doesn't say how many events they got or how good the picks were.
A Make Community member wanted quick AI answers without leaving Telegram, so he built a Make scenario behind a Telegram bot. Each new message goes through a router: typed text goes straight to the AI step, while a voice note is first fetched from Telegram in two steps (get the file path, then download the audio) and turned into text. Both end up in the same AI step, with a short system prompt so replies stay easy to read on a phone, and the answer goes back to the same chat. He says it has run since August, handles around 50 messages a day, and replies within a few seconds. His tips: the two-step voice download is easy to miss, and filters on the router made it much easier to debug.
The poster's OpenClaw work agent answers routine messages from coworkers, like where a document lives or how far along a project is, using his work tools. Everything else goes into a ranked list he works through, so he isn't pulled away by every ping.
The poster used a Claude Cowork inbox skill to archive mail and draft replies every hour. It sent one email without asking. No harm was done, but he disconnected Gmail and his other connections straight away.
While tidying up at night, the poster talked to ChatGPT Work by voice. It quizzed him on open items in his project notes, checked his email and Slack when he needed context, sent questions to teammates, and in the morning sent him his five most important tasks. He still checks emails with attachments himself before they go out.
Paste in today's emails and get back what needs a reply today, what can wait, what to ignore, and a one-line reply for each. It also flags emails that look like scams.