Stop asking which jobs AI will replace. Workday's Work Intelligence measures the tasks that change
Workday's early-access tool breaks jobs into tasks, links them to skills and tracks how work shifts with AI. The task is the right unit for planning. The risk is treating its output as decisions.
Listen to this article · 5 min
AI-generated narration of the full article.
“Which jobs will AI replace?” is the question boards ask, consultants answer and almost nobody can act on.
It is the wrong unit. A claims handler’s job is not one thing. It is reading documents, checking coverage, calling a customer who is upset, spotting the file that looks wrong, writing a decision letter and arguing with a colleague about an exception. AI changes some of those this year, some later and some never. A plan built on the job title averages all of that into a number that helps nobody.
Workday’s announcement on October 13 takes the other view. “AI rarely changes a whole job at once. It changes tasks, and as tasks change, the skills that matter most change, too.”
What Work Intelligence does
Work Intelligence, available in early access to Workday HCM customers, does four things according to the release:
- breaks jobs into the tasks they involve, which Workday says are often invisible in titles and org charts;
- connects those tasks to the skills they require and shows which skills are gaining importance;
- identifies who already has emerging skills and what training would close gaps, including for new roles such as managing agent performance;
- tracks over time how tasks change, how people’s time shifts and whether they move into new roles.
It builds on Workday Skills Cloud, which the company says more than 3,450 customers use, with an ontology of more than 75,000 skills. Those figures describe Skills Cloud, not Work Intelligence, which is new and in early access. The release’s subtitle promises to show leaders “where AI can help, what it should augment, and what has to stay human”; the body does not describe how specific recommendations are generated.
Why the task is the right unit
Task-level analysis fixes three problems in how enterprises plan AI today.
It finds the real candidates. Agent-ready work is rarely a whole role; it is a recurring task with clear inputs, a reviewable output and an existing owner, the tests we proposed in finding agent-ready work in existing processes. A task map makes those visible across the organization instead of in one workshop.
It makes value measurable. We argued that hours saved are capacity, not value: released time becomes value only when someone decides where it goes. Tracking how time shifts between tasks, and whether people move into new work, is the measurement that business cases usually skip. If Work Intelligence delivers that tracking, it addresses the weakest part of most AI programs.
It separates augmentation from automation. The same task can be automated for routine cases and augmented for exceptions. Planning at the task level lets that distinction exist on paper before it is designed into a system.
Where to be careful
A tool that recommends which tasks to automate is also, inevitably, a tool that shapes decisions about people. Three cautions follow.
A task inferred is not a task observed. If tasks and skills are inferred from job descriptions, profiles and system activity, the map reflects what the data shows, not necessarily what people do. Workers know which parts of their job are invisible to systems; they should validate the map before it informs anything.
A recommendation is not an employment decision. Treat the output as input to workforce planning, with human judgment and the usual consultation processes, not as a ranking of roles to cut. The ability to challenge the output is what makes a human review real, the gap IBM’s own research exposed in human oversight without judgment.
Vendor statistics are not evidence of results. The release cites Workday’s own workforce report (76% of heavy AI users expect new career opportunities from AI). That is a survey of expectations. Realized productivity is something each organization will have to measure for itself, which is, to be fair, what the tool is meant to help with.
A practical start, with or without the tool
Pick one function where AI adoption is already underway. List its twenty most frequent tasks with the people who do them. Mark each as automate, augment or keep, with a reason. Then measure, six months later, where the time went. Whether that list lives in Workday or a spreadsheet matters less than having it, because a plan that names tasks can be tested, and a plan that names jobs mostly cannot.