Should You Hire a Person or Deploy an AI Agent?

Next time you feel the urge to hire, pause and run the role through one question: how much of this job is describable process, and how much is judgment and relationships? The answer tells you whether to hire a person, deploy an agent, or, most often, do a bit of both. Here is a simple framework so you stop guessing and start deciding.

Here is the decision framework, the same one I use before any hire. It sits alongside the agent versus new hire cost comparison.

Step 1: List the tasks, not the title

"I need an admin" is a title, not a decision. Write down the actual tasks the role would do in a typical week: triage the inbox, schedule meetings, enter data, chase documents, liaise with clients, handle the odd tricky query. You cannot decide on a job; you can only decide on its tasks. This list is the whole basis of the framework.

Step 2: Mark each task process or judgment

Go down the list and label each task. Process means describable, repeatable, and rule-based, the kind of thing you could write instructions for. Judgment means it needs a human head: discretion, empathy, improvisation, relationships. Be honest, because the temptation is to over-label things as judgment to justify a hire. Most tasks are more describable than they first feel.

Step 3: Add up the split

Now look at the balance. If the role is, say, 80% process and 20% judgment, an agent handles the bulk and you barely need a person. If it is 30% process and 70% judgment, hire the human, an agent would only nibble the edges. Most roles land somewhere in the middle, which points to the blend rather than an either-or.

Step 4: Compare the real cost

Cost the two options honestly. A human is not the headline salary; it is salary plus tax, pension, tools, software, workspace, and your management time, the fully-loaded figure. An agent is a few hundred euro a month plus a few minutes a day of your approval. Weigh the fully-loaded human against the agent-plus-review, then factor in the hours reclaimed and any revenue unlocked, laid out in the ROI of AI agents.

The decision table

A tangible example: imagine Ruth's dilemma

Imagine Ruth, about to post an ad for a €3,000-a-month operations hire. She runs the framework and finds the role is roughly 65% process, scheduling, data, chasing, reporting, and 35% judgment and client contact. Instead of one full hire, she deploys an agent for the process at a few hundred a month and brings in a part-time person for the judgment 35%. It is easy to picture Ruth getting the role covered for less, with the human doing more interesting work than pure operations ever offered. The framework turned a reflexive hire into a sharper decision.

When to just hire the human

The framework is not a machine that always says automate. Sometimes the honest answer is a person, full stop: when the role is mostly judgment, when relationships are the point, when the work will grow beyond anything you could write down, or when you simply need a human presence to think, care, and be accountable. Forcing an agent into those roles damages the very thing that made them valuable. A framework that always says automate is a sales pitch, not a tool.

When to just deploy the agent

Equally, some roles are so describable that hiring a person for them is spending money to create management overhead. Pure data entry, first-line triage, scheduled chasing, routine reporting, if that is the whole job, an agent does it faster, cheaper, and without the recruitment, onboarding, and holiday cover a hire brings. Deploying the agent frees the budget you would have spent on drudgery for something that actually needs a person.

Watch out for the false economy on both sides

The framework can be misused in two opposite ways, and both cost you. On one side, owners over-automate, forcing an agent into a judgment-heavy role because it is cheaper, then wondering why clients drift away or quality slips, the saving was false because it hollowed out the part that mattered. On the other, owners over-hire, taking on a full person for work that is almost entirely describable, then paying a salary plus management overhead for drudgery an agent would run for a fraction. The framework guards against both only if you are honest at step two about which tasks are genuinely judgment and which merely feel that way. Honesty there is what keeps the decision sound.

Revisit the decision as the role grows

A role is not fixed forever, so treat the hire-or-automate call as something you revisit rather than settle once. A job that starts almost entirely describable may grow a judgment-heavy edge as your business matures, at which point you add a human on top of the agent already running the process. Equally, a judgment role may accumulate routine tasks worth handing to an agent over time. Running the framework again every so often, rather than assuming last year's answer still holds, keeps each role matched to what it has actually become.