The public conversation about AI and work tends to jump from automation to displacement. The operating reality is more complicated. Most organizations will not replace whole jobs first. They will reallocate tasks, change the quality bar, compress cycle times, and ask managers to decide which work should be automated, augmented, or left alone.

OECD employment and skills work points to a persistent theme: exposure does not equal outcome. The outcome depends on institutions, training, job design, bargaining power, and how quickly organizations can turn technology into better work rather than just cheaper work.

The task is the unit of change

Workforce planning still happens at the role level, but AI arrives at the task level. A finance analyst does not become obsolete all at once. Forecast commentary, variance explanation, invoice matching, and scenario drafting each change differently. The useful workforce plan starts there.

Exhibit 1
Task exposure heatmap
Where AI changes work first
Drafting and synthesis
High
Pattern detection
High
Routine decision support
Rising
Stakeholder judgment
Human-led
Physical service delivery
Low
Source: Kaya Development synthesis of OECD employment and skills research and public labour-market evidence

Reskilling needs a portfolio

Most reskilling programs are too broad. They train everyone on the same tool and hope productivity appears. The stronger approach is a portfolio: protect critical roles, redesign high-exposure tasks, train managers, create expert pathways, and build internal mobility for roles whose work will shrink.

Exhibit 2
Reskilling portfolio
Where training investment should go
Near term
Task redesign
Manager training
Tool fluency
Risk controls
Two years
Internal mobility
Data skills
Expert pathways
Job redesign
Source: Kaya Development workforce transformation benchmarks

The middle layer is decisive

AI changes management work before it changes executive work. Middle managers decide which outputs are acceptable, which exceptions matter, and which old processes can be retired. If they are not trained to redesign work, AI becomes another tool layered on top of already crowded workflows.

Exhibit 3
Manager-role redesign funnel
How AI changes the middle layer
Map team tasksWhat work is repetitive, judgment-heavy, or risky?
See
Classify automation potentialAutomate, augment, protect, stop
Choose
Redesign workflowControls, handoffs, quality bar
Build
Coach adoptionTrust, usage, feedback loops
Lead
Retire old workRemove duplicate manual process
Capture value
Source: Kaya Development AI workforce operating model

What leaders should do next

Do not start with a headcount number. Start with a task map for the five functions where AI exposure and business value are highest. Then train managers to redesign work, not just to use tools.

Three moves now
  1. Map tasks before roles.
  2. Train managers as work designers.
  3. Make internal mobility part of the AI business case.