Comparison 06 · stated plainly, no scorecard
Most of these are good. They stop earlier.
Comparison pages that give every rival a row of crosses are worth nothing to a buyer. Here is what each of these actually is, how it is sold, and the point at which it hands the work back to you.
| What it is | Where it runs | How it is sold | Where it stops |
|---|---|---|---|
| A desktop agent | Your laptop | Metered, one vendor's models | It works on your machine, with your files, and it stops when your machine does. Excellent for the hour you are at the desk. |
| An in-app copilot | Inside the suite | Per seat, per month | It helps with the document that is open. Ask it to own a process that spans six systems and it is not that product. |
| An enterprise chat account | The vendor's cloud | Per seat, with usage caps | A very good conversation, one vendor's models, and your data in their estate. Long unattended work is not what a seat licence prices. |
| A CRM or service agent | The CRM | Per action, plus licences | Strong inside its own object model. The work that crosses finance, engineering and the inbox is outside it. |
| A model platform | Wherever you deploy it | Per token | You get a model, which is the easy half. Orchestration, memory, recovery and approvals are yours to build and to keep building. |
| A general agent | The vendor's cloud | Subscription | Impressive on a self-contained task. The estate integrations, the deployment boundary and the audit trail are where enterprise buying starts. |
| Ninja | Your cloud, or ours | Credits, or capacity you own | It keeps the work for hours or weeks, on the models you pick, inside the boundary you drew, and it tells you what it did. |
Categories, not a takedown. Every one of these is a real product with real users, and for a lot of jobs one of them is the right answer.
Ask these
before you
ask about
features.
How long can one piece of work run without a person?
If the answer is measured in minutes, you are buying an assistant. That may be exactly what you want, and it is a different budget.
Whose machines is it on when the data is sensitive?
The answer decides whether your security review is a conversation or a year.
Does the price go up when it works harder?
If the meter runs on every step, somebody in finance will eventually ask the team to use it less. That is the moment the deployment dies.
Not every job
needs a run.
If what you want is a better autocomplete inside your editor, buy the copilot. If you want to talk to a model about a document, an enterprise chat account is cheaper and quicker to roll out. If you have a strong platform team and a very specific pipeline, take a model and build it.
Ninja earns its place when the work is long, spans several systems, has to keep going while nobody watches, and cannot leave your estate. That is a narrower claim than most of this category makes, and it is the one worth testing.
Everything on this page describes categories of product as they are publicly sold. No competitor is named, no benchmark is claimed, and nothing here should be read as a statement about a named vendor's roadmap.