ChatGPT Work vs. Claude Cowork: what's actually new and what Cowork already solved

A few days ago, Spanish content creator Gustavo Entrala published a video demoing the new ChatGPT desktop version live, comparing it to leaps like the arrival of the iPhone — and pointing out that it's inspired by a product Anthropic shipped months earlier: Claude Cowork. It's worth a point-by-point comparison of how different it really is from what Cowork already does.

Voice and computer control

In the demo, Entrala asks ChatGPT by voice to review an email and leave a ready-to-send reply in Gmail without touching the keyboard. Cowork doesn't yet have an equivalent native voice mode, but it handles computer control with tiered permissions instead of full access: a browser can be viewed but not clicked without explicitly enabling it, a terminal runs specific commands without receiving free-form keystrokes, and any irreversible action requires explicit confirmation before it happens.

Emails, recurring tasks, and documents

Leaving an email ready as a draft, scheduling a daily news digest, or generating a presentation from already-researched content — everything the ChatGPT demo shows has a direct equivalent in Cowork: its Gmail connector leaves replies as drafts pending explicit approval, its native scheduled tasks cover both one-off reminders and recurring routines, and it generates real Word, PowerPoint, Excel, and PDF files — not plain text inside the chat.

Data analysis and what Cowork doesn't do yet

Connecting a data source (analytics, a CRM, a database) and asking it direct business questions is the same principle behind Cowork's connectors. Where there's a real gap is advanced creative control: the color grading inside Final Cut Pro shown in the demo is territory where Cowork is more conservative, prioritizing web browsing and office workflows over professional editing.

Want this for your company?

The underlying difference

Beyond the feature checklist, the difference is philosophical. Entrala frames ChatGPT's leap as a matter of capability; Cowork was built by first solving how safe it is to hand these tasks to an AI, building permission, confirmation, and limits into the design from the start. At Kempro we run into that same question every time we advise a client: it's never just how powerful the technology is, but how well its permission model fits the level of governance the organization actually has.

Both follow a tiered subscription model — it's worth confirming current pricing on each company's site. But Entrala's video confirms something we probably already suspected: the "chatbot that answers questions" is behind us. The conversation now is about how much control we hand an agent over our work tools.