Five products now do roughly the same thing: an agent that keeps its instructions, works across sessions, and does multi step jobs while you are elsewhere. Grok Bot, ChatGPT Work, Claude Cowork, Gemini Spark and Manus. They are not interchangeable, and the differences are structural rather than a matter of which model is better this month.
What each is built around
Grok Bot
A team of named agents, each with its own description, its own enabled skills and its own routines, running on their own cloud computer. The distinguishing pieces are the routine, which lets a Bot work on a schedule without being asked, and share as template, which turns a configuration into a link somebody else can add to their account in one click. That second one is why a template library like this one is possible at all: a Grok Bot is a portable artefact rather than a private setting.
ChatGPT Work
Persistent agent work inside ChatGPT, which for most people is the lowest friction way to try the idea because the account already exists. It is the strongest option if the work you want done lives in the same place your chat history already does.
Claude Cowork
Built for desktop knowledge work: local files and selected applications on the machine in front of you. If the job is a folder of documents on your own disk rather than a browser tab, that is a genuine architectural difference rather than a preference.
Gemini Spark
The option that makes sense if the work is already inside Google Workspace, for the same reason ChatGPT Work makes sense if it is already inside ChatGPT. Proximity to the data beats capability more often than people expect.
Manus
Long running cloud tasks that produce a deliverable, in a managed workspace. Closer to commissioning a piece of work than to having an assistant on hand.
The question that actually decides it
Not which model is best. Where does the work live?
- Files on your own machine, and the agent needs to see them: Claude Cowork.
- Inside Google Workspace: Gemini Spark.
- Inside ChatGPT already, and you want the shortest path to trying it: ChatGPT Work.
- A long task you want to come back to finished: Manus.
- A set of named specialists you can schedule, share and hand to somebody else: Grok Bot.
Where Grok Bot is genuinely different
Two things, and both are about the configuration rather than the model.
Routines. A Bot that runs at seven every weekday is a different kind of tool from one you have to open. It changes what you use it for: a daily briefing or a weekly competitor check are only worth having if they arrive without being asked.
Portability. Because a Bot can be shared as a template and copied to another account, a good configuration is transferable. That is why the specialists are worth writing carefully once: a code review Bot that ranks by blast radius can go to the whole team as a link rather than as a document somebody has to reimplement.
A note on pricing
All five sit on subscription tiers that have changed more than once this year, and the plan a given agent product appears on is the part that moves most. Check each vendor's own pricing page rather than a comparison table, including ours: any figure written into a guide in August is a figure somebody reads in March.
If you are trying Grok Bot
Start with one Bot doing one job you would otherwise do yourself in under ten minutes, then read how to create a Grok Bot for the setup and how to write instructions that change the output for the part that decides whether it is any good. Or skip the writing and take a finished one.
Whichever you pick, the design still has to exist
None of these five products removes the need for a screen that looks finished, and all of them build faster from a reference than from a blank prompt. The design templates here work with all of them, grouped by the tool you build in: Lovable, Bolt, Replit, Claude Code, ChatGPT Codex and Cursor templates.