AI Reveals Three Task Tiers Inside Job Descriptions

For nearly 25 years, I have watched job descriptions present a tidy fiction: that a role is a stable bundle of tasks, performed the same way, forever. That fiction has become unsustainable. Artificial intelligence has fractured the single job title into three distinct categories of work — tasks best done by machines, tasks that require human and machine collaboration, and tasks that must remain entirely human. When an organisation treats a role as one undifferentiated unit, it risks measuring, hiring for and rewarding the wrong things in at least two of those three categories.

The three tiers of human-AI synergy in job tasks

Before rethinking a single job description, the question must shift from “what does this role do?” to “what is the actual operating relationship between the human and the AI agents, task by task?” The most practical framework for answering that question breaks work into three tiers.

The future of role design is not about how much AI a role uses.

Uniquely human tasks

These are the parts of any role built on relational trust, empathy and moral judgment. They involve daily decisions that carry real consequences for another person — a team member, a colleague or a client. AI can inform this work by providing data summaries and context, but it cannot carry accountability for it. Examples include redundancy conversations, disclosures about health and wellbeing, and outcomes of employee relations incidents. When a manager decides to support an internal promotion even though the numbers alone do not reflect the person’s full value, that is where human judgment and empathy matter most.

AI augmented tasks

This is the truly symbiotic zone where machines and humans merge on a task. AI might generate scenarios or synthesise data, while humans provide framing, contextual judgment and final sign-off. Most knowledge work is moving rapidly into this space. The skill required is not simply “can you use this AI tool?” but “can you tell when the tool’s output is subtly incorrect, can you push it toward a better answer, and can you take responsibility for what you ship?” That competency is rarely tested during hiring or measured for performance and talent management.

AI automated tasks

Scheduling, tracking, basic reporting, collating and summarising large volumes of information can now shift to AI agents. The human role moves from performing those tasks to overseeing the AI agents that have automated repetitive but important work. In these cases, high performance is no longer about doing the work manually and doing it well. The critical skill becomes spotting when the automated output is wrong.

EEOC Sues Bollinger Shipyards over Suboxone Ban for Safety-Sensitive

Employers drop barriers for 10 million disabled workers

Long hours and compliance push 44% of transport workers to exit

Why AI enablement breaks traditional job evaluation

Job architecture has historically been built role by role and level by level, as though a “senior” version of a job is simply a more complex version of the same tasks. Traditional job evaluation tools — such as Paterson, TASK and the WTW factor set — have graded roles on a handful of differentiators: how far ahead planning extends and how long before results appear, whether the work is repetitive or demands judgment in the moment, and whether communication is meant to inform or to persuade someone more senior or external.

AI enablement does not let those differentiators sit neatly at the level of a whole role or grade. Instead, they now vary at the level of a single task inside a role. A task that is repetitive, plannable and short-horizon is precisely the profile AI absorbs first, so it slides into the AI-automated tier, and the human’s job becomes governing the output rather than performing the task. A task where the impact takes months to surface, where full planning is impossible because the situation has not yet occurred or too many variables exist, ends up in the AI-augmented or uniquely human territory — because judgment under genuine uncertainty is something AI can neither perform nor be accountable for.

Persuasion that gets a sceptical board or external stakeholder to act cannot be automated at all. AI can help draft the argument, but the credibility that makes someone believe and change their mind depends entirely on the human delivering it and their relational abilities.

An uncomfortable implication for job grades

Scoring a whole role against these traditional factors is starting to miss what is actually happening. Two people on the same job grade, evaluated the same way, can now be doing genuinely different work because their tasks have split differently across the taxonomy, even though the evaluation score says they are equivalent.

Consider recruitment as a practical example. Sourcing and CV screening is increasingly AI-automated — useful for high-volume recruitment, but still requiring human governance. A recruiter’s value shifts to catching when the applicant tracking system has screened out a good candidate for a bad reason. Interview calibration falls into the AI-augmented tier: AI can draft structured scorecards and flag inconsistent ratings across a panel, but a human still needs to weigh fit, potential and risk in a way that resists reduction to a simple numerical score. The actual conversation in which someone receives an employment offer or is told they were unsuccessful after three interview rounds is uniquely human.

The employee engagement consequences

Gallup’s research has shown for years that people disengage fastest when they cannot see how their actual contributions connect to what gets recognised and rewarded. Organisations are layering AI into roles without redefining what “good” looks like at each tier, creating that mismatch at scale and pace. The same recruiter may be punished for slower screening even when slowing down was the right call. A manager may be rewarded for staying hands-on with admin and reporting tasks that should have been automated months ago. Current engagement questionnaires are unlikely to pinpoint this as a reason for low morale.

How organisations are starting to respond

A full taxonomy rollout across an entire organisation is not necessary to begin. The author suggests starting with one role — even one’s own — and mapping all tasks and responsibilities across the three tiers. That quickly reveals which parts of the role are being measured on the wrong basis. It also tends to show where the best people are quietly burning out because they are doing work that should have been automated, or they are “bot-sitting” and correcting questionable AI output.

Some of the early observations from organisations that have tried this approach include:

• Taking one HR role and breaking it into tasks before scaling the exercise to other functions.
• Sorting the tasks into the three categories — uniquely human, AI augmented, AI automated.
• Examining how performance is measured; if someone is being judged on speed for work that demands careful judgment, that is worth addressing first.
• Shifting hiring criteria toward the ability to evaluate AI output rather than simply the ability to use a tool.
• Questioning whether existing grade levels still make sense when two people on the same grade may now be doing very different jobs.

The future of role design is not about how much AI a role uses. It is about knowing, task by task, which parts of the work were never meant to be automated in the first place — and then figuring out how to recognise and reward people for their uniquely human contributions. That rethinking begins with a single job description, broken open.