How to Update Job Descriptions for AI

Whose job is it to redefine roles for AI, HR or the manager?

Both, together. HR holds the framework and the fairness lens. The manager holds the ground truth about what the work requires day to day, and a role redesign without that input is a guess written up as policy.

HR is often asked to lead this because job descriptions are an HR document. But HR rarely sits close enough to the daily workflow to know which parts of a role have shifted and which have only had AI added. The manager watching the team use AI, or avoid it, every day has that information.

This is where ground truth before prescription matters. You cannot prescribe a new role definition before you understand how the work happens on the ground, including where people have built workarounds or ignored AI entirely. A manager who sees a team member still doing a task by hand, with AI available for it, is holding information that should shape the job description. It is rarely a performance problem.

The practical fix is joint ownership. HR sets the structure, keeps roles fair and consistent, and keeps the language usable for compensation and hiring. The manager supplies the specifics: what changed, what didn't, what the team is pushing back on and why. Leave either one out and you get a document that is legally tidy and operationally wrong, or accurate on the ground and inconsistent across the organization.

Why do updated role expectations not change how people work?

Role language changes behavior when it comes with changed decision rights, changed quality standards and some of the old workload taken away. Without those, the update describes an intention and leaves the job as it was.

The common version: the job description is rewritten, and the old job stays fully intact underneath it. Nothing came off the person's plate. They are expected to deliver at the same pace and quality, and also to learn AI, rethink how they approach their own work, and get good at all of it quickly. That is an unfunded mandate with a new paragraph attached.

People notice the distance between the words and the reality faster than leadership does. If the document says "uses AI to accelerate creative development" but the performance review still measures the same output volume on the same timeline, with no allowance for the learning curve, the sensible move for the employee is to ignore the new language and protect their numbers the old way. That is a reasonable response to a document that asked for new behavior without changing the conditions that drive the old behavior.

The step most rewrites skip

Few role updates name what gets removed. If AI now drafts the first version of a report, the job description should say the person no longer spends hours on first drafts, and should say what they do with the recovered time. Skip that step and the role becomes two jobs stacked on top of each other.

What should change in a role beyond the job description when AI is introduced?

Decision rights, quality standards and the workflow itself: who approves what, what "good" looks like now, and which steps moved, disappeared or got added. The document is the last thing to update.

Decision rights matter because AI often produces options faster than a person can evaluate them, so someone now has to decide what is good enough to move forward, and that approval step may not have existed in the old workflow. If nobody has been given that authority, it either bottlenecks at a manager who didn't ask for the extra work, or it gets skipped and quality slips.

Quality standards need the same scrutiny. A role measured on output volume, how many briefs written, how many drafts produced, usually needs a different measure once AI can generate volume on demand. The valuable part of the job shifts toward judgment, taste and knowledge of the customer, which an AI model does not supply on its own. If the performance standard does not move with the work, people keep optimizing for the old measure, because that is what gets rewarded.

This is also where many organizations find they are redesigning more than one role at a time. A team's workflow rarely changes in isolation, so updating one job description without looking at the roles on either side of it tends to create friction somewhere else in the chain.

How do you know if your AI role updates are working?

You will see it in behavior first: people using AI without being reminded, decisions being made at the level they are now supposed to be made at, and the old manual version of the task disappearing from someone's week instead of running alongside the new one.

A role update that is working reduces friction in the handoffs around that role. If a manager used to spend time reviewing first drafts and now spends it on something else, the redesign took. If the manager reviews drafts at the same pace and now calls them "AI-assisted drafts," nothing structural changed.

Watch for the opposite signal too: people building workarounds, reverting to old methods, or doing the AI-assisted version of a task alongside the manual one just in case. That is data. It usually means the update asked for new behavior without removing the old expectation that made the new behavior feel risky.

When the document is honest and the organization isn't

A job description can say all the right things about AI and still misdescribe the role if the workload, timeline and support have not moved to match. People can tell a role that changed from a role that got new language. The Elephant in the Algorithm looks at why that gap opens, including what happens when leaders announce an AI-first posture before the underlying work has changed.

If you already know the roles need rebuilding and want a clear plan for doing it with your managers, see how the AI Profit Partnership works. It starts by researching how your team works today, puts quick wins in place along the way, and ends with a plan for the year ahead.