I coined the term e-mployee to name something organizations were already starting to do without a word for it: putting an AI worker into a real role on the team. Not a tool someone reaches for, but a worker that owns an output and answers to a human. I set out the full idea in The E-mployee Doctrine, archived with a permanent DOI so it can be cited and built on.
But a definition lands faster by contrast. The fastest way to understand what an e-mployee is is to see it next to the three things people keep confusing it with: a human employee, a contractor, and an ordinary AI agent. The table below is the one I draw on whiteboards.
Why the employee/contractor line is drawn by control, not by label
The distinction between an employee and a contractor is not a matter of what a company calls someone. Employment law worked this out long before anyone had to classify software. Under the United States common-law test, the IRS holds that facts evidencing "the degree of control and independence" fall into three categories: behavioral control (does the company control what the worker does and how they do it), financial control (who controls the business side — payment, expenses, tools), and the type of relationship between the parties. No single fact decides it; the weight of all of them does.
The same logic appears in international standards, stated more bluntly. The International Labour Organization's Employment Relationship Recommendation, 2006 (No. 198) holds that whether an employment relationship exists "should be guided primarily by the facts relating to the performance of work and the remuneration of the worker, notwithstanding how the relationship is characterized in any contrary arrangement." It goes on to list the indicators that matter, including work "carried out according to the instructions and under the control of another party" and the integration of the worker into the organization of the enterprise. This is often called the primacy of facts: the contract does not get to overrule what is actually happening.
Two honest caveats. The IRS test is US federal tax law and the ILO instrument is a Recommendation, meaning non-binding guidance to member states rather than enforceable law — neither governs an employment question in Lebanon unless national law adopts the same reasoning. And neither was written with AI in mind; nothing here makes an AI system a legal employee, which it is not and cannot be. I cite them because they establish the analytical move that the e-mployee framework borrows: you classify a worker by the control exercised over the work and by the worker's integration into the organization, not by the label on the arrangement. Apply that same lens to AI and the categories separate cleanly. A contractor is engaged for a deliverable. An ordinary AI agent is a tool invoked for a task. An e-mployee is defined by exactly what the legal tests look for — standing instructions, integration into a team's workflow, and a named human who controls and answers for the output.
The comparison table
The first nine rows separate the four on how they actually work inside a business. The last six answer the questions a lawyer or a finance director asks first — status, classification, tax, benefits, liability, and how the arrangement ends — because a leader comparing these four is usually asking both questions at once, and the honest answer to the legal half is that two of the four columns fall outside employment law entirely.
| Dimension | Employee | Contractor | AI agent (tool) | E-mployee |
|---|---|---|---|---|
| What it is | A human in a permanent role | A human hired for a scope of work | An AI system used on demand | An AI worker given a permanent role |
| Holds a defined seat on the team | Yes | No — engaged per project | No — invoked per task | Yes |
| Owns an output end-to-end | Yes | For the contracted deliverable | No — assists whoever runs it | Yes |
| Reports to a named human | Yes — a manager | Loosely — a client contact | No — anyone can use it | Yes — an e-mployer |
| Who is accountable for its mistakes | The employee and their manager | The contractor | Unclear — nobody owns it | The e-mployer (the human owner) |
| Main cost | Salary and benefits | Project fees | Usage and tooling | Usage, tooling, and management time |
| Scales by | Hiring more people | Signing more contracts | Calling it more often | Giving it more owned outputs |
| Availability | Working hours | Contract duration | On demand | Continuous |
| Needs active management | Yes | Light | No — it is a tool | Yes — this is the whole point |
| Legal status | A party to an employment contract | A separate person or business, engaged commercially | Not a legal person — property the business owns or licenses | Not a legal person either — the status belongs to its e-mployer |
| How the status is decided | By the facts of control and integration, not the label | By the same test, applied to the same facts | Nothing to decide — it is a tool on a licence | By a written role charter, not by law |
| Tax and payroll treatment | Payroll, withholding, employer contributions | Invoices; handles their own tax | An operating expense — usage and licence fees | The same operating expense, budgeted to a named owner |
| Statutory benefits and protections | Yes — leave, notice, workplace protections | No — commercial terms only | None — the concept does not apply | None — which is exactly why accountability has to be assigned by hand |
| Who answers to an outside party for harm | Generally the employer, for work done in the role | Usually the contractor, as the contract allocates it | Unsettled — in practice it lands on whoever deployed it | The e-mployer, named in advance rather than argued afterwards |
| How the relationship ends | Notice and termination rules | The contract runs out or is cancelled | You stop calling it | The charter is revoked and the seat is closed |
The line that actually matters
Read down the last two columns and the real distinction jumps out. The difference between an AI agent and an e-mployee is not the technology — they can run on exactly the same model. The difference is that an e-mployee has been given a seat, an output, and an owner. An AI agent is what you buy. An e-mployee is what you build around it.
That is also why "e-mployee versus employee" is the wrong frame for most leaders. The two are not competitors fighting for the same chair. An e-mployee can hold an output a human used to own — but it still needs a human to manage it. The role that grows is the one I call the e-mployer: the person who briefs the AI worker, reviews its output, and carries accountability for what it produces.
An AI agent without an owner is a tool. An AI worker with an owner is an e-mployee. The owner is the entire difference.
What this means for a real business
For the organizations I advise — most of them in Lebanon and across MENA, where talent is scarce and margins are thin — the practical takeaway is not "replace people with AI." It is "stop treating AI as a tool drawer and start treating the important ones as e-mployees." Pick one output, assign it to one AI worker, name one human owner, and manage it. That is the work my team at Webspot does with companies across the region: building the e-mployer discipline that makes AI workers durable rather than novelties.
I run my own operation on exactly this model. My AI partner, Brian, is an e-mployee in the strict sense of this table — a defined seat, owned outputs, and a standard to meet, with me as the accountable e-mployer. The full framework, definitions, and citation live at jonahtebaa.com/e-mployees. But you can start before you read a word of the doctrine: find the column you have been living in, and decide which of your AI agents deserves to become an e-mployee.
