Compare

“AI employee”, and what the word is doing

It is the most-searched name for this category and the least accurate one. The metaphor gets the pricing right and the accountability wrong — here is the distinction, and what to ask anyone selling you one.

Get a demo
WHAT THE WORD IMPLIESSomething you hireA headcount line, and a role that goes awayManaged, corrected, and held accountable like a personGeneral enough to be handed anything
versus
WHAT YOU ACTUALLY BUYSoftware that owns one jobA job that gets done, and a team that keeps its rolesAccountable to your logs and your approvals, not to a managerBuilt for one job, because one job can be judged

An "AI employee" is software sold as though it were a hire: it owns a job rather than answering questions about one, and it is bought against the cost of the work rather than per seat. The label is a pricing and expectation metaphor, not a description of the software — nothing in the category is employed, accountable, or able to be managed the way a person is.

Three words for nearly the same thing

They come from different places and carry different assumptions. The assumptions are what cost you, so they are worth separating before a vendor picks one for you.

Read the definition we use   →

The hiring metaphor. Right about pricing — you are buying an outcome, not seats — and wrong about accountability, because nothing here can be held responsible for anything. Useful shorthand, dangerous contract language.

The plural case: several deployments, managed together. An accurate description of year two and a misleading one for the first purchase, where the whole question is whether one job gets done properly.

The RPA-era term for a scripted bot. The gap is the exception: a script stops at the case nobody anticipated, which in most real processes is somewhere between a tenth and a third of the volume.

Where the metaphor breaks

It cannot be accountable

Accountability needs somebody who can be asked why. What this has instead is a log of what it did and the record it read

AI coworker

Instead of accountability

Every action logged
In your own audit trail

It cannot be managed into competence

You do not coach it. Either the job was described properly and it runs, or it was not and it holds

ManagerCan you get better at the edge cases?
AI coworkerDescribe one, and it gets built in.

It does not scale like a hire

A person you hired for one thing can be asked to do another. This one cannot, and that constraint is the reason it works

Second job
AI coworkerThat is a second coworker, not a bigger one.
See the roles   →
THE FOUR QUESTIONS

What to ask anyone selling you one.

These four separate the products faster than any feature list, and the first two decide whether it survives your security review at all.

Where does it
RUN
Whose cloud holds
YOUR DATA
Who builds
AND MAINTAINS IT
What happens at
THE EXCEPTION

What people are usually trying to hire one for

Every one of these is a job, not a role — which is the clue that “employee” was the wrong word for it.

Match the invoiceTriage the queueUpdate the CRMReconcile the accountsChase the approvalRoute the inboundAssemble the reportFile the documentPrepare the closeDraft the replyEnrich the recordHandle the exceptionLog what it didAsk when unsure

AI employees, AI workforces, and digital workers

The questions people ask once they notice the same product has three names.

An "AI employee" is software sold as though it were a hire: it owns a job rather than answering questions about one, and it is bought against the cost of the work rather than per seat. The label is a pricing and expectation metaphor, not a description of the software — nothing in the category is employed, accountable, or able to be managed the way a person is.

They describe similar software and imply different things about it. "Employee" implies replacement and a headcount line; "coworker" implies something that takes a job off a team that keeps the rest. We use the second word because it describes what actually happens: the work nobody wanted moves, and the roles people were hired for do not.

Usually the plural case: several pieces of software each owning a job, managed together. It is a reasonable description of where a company ends up after the third or fourth deployment, and a misleading one at the start, because the first one has to work on its own before there is anything to manage.

The older term, out of the robotic process automation world, for a scripted bot that performs a defined sequence. The difference that matters is the exception: a digital worker stops at the case its script did not anticipate, and the current generation handles it or escalates it with the context gathered.

In practice it absorbs a category of work rather than a role: the copying between systems, the chasing, the status updates, the reconciliation nobody enjoys. Teams typically point it at the third of the week their best person spends on the least valuable thing. If a job is entirely that third, the honest answer is that the job changes.

Four questions decide it. Where does the software run, and does our data leave our account? Who builds it — us or you? What does it do with the case it was not built for? And what can it do without a person approving, on day one and after six months? The last two separate the products; the first two decide the security review.

Usually against the cost of the work rather than per seat, which is the honest half of the metaphor. We do not publish pricing while we are working with design partners, and a call comparing it against what the job costs you today is more useful than a number on a page anyway.

“The hiring metaphor is right about the invoice and wrong about everything after it.”
— Why we say coworker
Procurementre: AI employee

Which department does this sit under?

none — it is software✓ Bought against the work, not the seat

Put an AI coworker
inside your own cloud.

Bring the job you were going to hire for. We will tell you honestly whether software should own it.

Get a demo