What is the plain definition of a digital worker?
A digital worker is software that performs the work of a role end to end under human oversight. It is given a remit, connects to the systems where that work happens, produces the output the role is responsible for, and reports to a person.
The phrase is deliberately job-shaped rather than technology-shaped. It says nothing about how the software works internally, which is both its strength as a category label and the reason two products sold under it can be radically different in what they can actually handle.
Where did the term digital worker come from?
The term digital worker was popularised by the robotic process automation industry, well before the current generation of language models. In that era a digital worker was a configured bot that followed a recorded sequence of steps across applications: open this system, read this field, enter it there, repeat.
That history matters when evaluating anything sold as a digital worker today, because the label carries forward while the underlying capability has changed completely. A buyer who assumes the modern meaning and gets the older architecture will be surprised by how brittle it is; a buyer who assumes the older meaning may dismiss a system that can genuinely reason.
What separates a modern digital worker from a scripted one?
A modern digital worker reasons over evidence rather than following a recorded script, which is what lets it handle cases nobody anticipated. Given an unfamiliar invoice format, a scripted automation fails or, worse, silently mis-reads it. A reasoning system reads the document, decides what it is looking at, and raises an exception when it is unsure.
This is why exception handling is the sharpest evaluation question. Scripted automation is excellent when the process is stable and the inputs are uniform, and it degrades badly at the edges. The edges are usually where the human time actually goes.
How does a digital worker differ from a Digital FTE?
Digital worker and Digital FTE describe the same category with different emphasis. Digital worker is the neutral, technology-agnostic label. Digital FTE borrows the workforce-planning unit for full-time equivalent and therefore makes a stronger claim: that the software covers a role's capacity and is managed like one.
In practice, treat Digital FTE as digital worker plus three commitments: a named role with targets, an approval gate on outward actions, and a management rhythm you can see. Those commitments are testable during an evaluation, which is why the narrower term is more useful when comparing vendors.
What should you ask when evaluating a digital worker?
Evaluate a digital worker on what happens when the input is unusual, because that is where the categories diverge and where the value sits. A demo on clean data tells you almost nothing; every product handles the happy path.
Bring a genuinely awkward example from your own operation: the invoice with the odd layout, the ticket that spans two policies, the lead with a broken record. Watch whether the system recognises that it is uncertain, and what it does about that uncertainty. A system that escalates with a clear reason is safer to deploy than one that always produces an answer.
- What happens with an input format it has never seen?
- How does it signal uncertainty, and to whom?
- Which actions require a human approval, and can we change that list?
- Where is the record of what it did and why?
Who manages a digital worker day to day?
A digital worker is managed by the person who owns the function it works in, not by the IT team that connected it. Placing ownership with the function is what keeps the remit and targets honest, because that person feels the consequences of both.
IT and security own the access: which systems it can reach, with what permissions, and how those are revoked. That split — function owns the work, IT owns the access — is the arrangement that tends to survive contact with reality, because each group is accountable for the thing it can actually control.
How does a digital worker earn more responsibility over time?
A modern digital worker earns responsibility the way a person does: from a record. On ftes.ai every Digital FTE starts at Level 0 on the autonomy ladder, where every action is a proposal that waits for a named person. When a specific class of action, such as exact-match reconciliations or interview scheduling, shows a long run of approvals with no edits, its manager can promote that class to auto-approve within stated bounds, leaving every other class gated. The promotion is logged with the manager's name and the record it rests on.
Responsibility is also taken back the same way. A send-back in a promoted class, a change in the sources the class relies on, an incident, or a new manager are each reasons to step it back, and doing so is one logged change that takes effect immediately. A digital worker whose responsibility cannot be granted per class from evidence and withdrawn in one action is a script with a trust setting, and the difference matters most in the months after the first week, when the novelty has gone and the work is simply being done.