Organizational development leadership for AI-enabled work

Define the organizational decision

Start with one workflow and one decision such as whether to change responsibilities, create a new review step, redesign a team interface, add practice, or continue an experiment. Describe the current output, users, sources, handoffs, delays, quality checks, exceptions, and accountable owner before proposing a future state.

Classify the organizational development leader as owner, reviewer, adviser, or collaborator for each element. They may own facilitation, role design, change learning, and feedback. Functional leaders own work quality; managers allocate work; technology operates systems; legal, privacy, security, and employee-representative bodies retain their responsibilities.

The OECD's 2026 report Building an AI-ready public workforce links workforce development with institutional strategy, governance, multidisciplinary capability, experimentation, and continuous learning. Its public-sector evidence can inform design questions but does not prescribe one corporate structure.

Sources: [1]

Build the work-system transition map

Create one row per material task or handoff. Record the current actor, decision, inputs, output, source authority, AI contribution, retained human judgment, future actor, required capability, review capacity, changed incentive, accessibility need, failure path, escalation, evidence to collect, owner, and review date.

Separate task removal, task reduction, task creation, and task displacement. An assistant may reduce first-draft effort while creating source checking, exception routing, or stakeholder explanation. Name where that work lands and whether the receiving role has time and authority to perform it.

Include the transition path, not only the proposed end state. Record which responsibilities change first, which teams need practice, what temporary coordination is required, and how the organization will detect overload or ambiguity before expanding the design.

The UK government's 2026 employer guide to AI upskilling emphasizes practical, reachable, integrated, modular, expandable, and sustainable learning. It supports linking capability building to real roles and workflows, while local transfer and organizational outcomes still require direct evidence.

Sources: [2]

Worked artifact: a synthetic proposal-review team

Imagine a synthetic proposal team introducing an assistant that extracts requirements and drafts a compliance matrix. The transition map shows that coordinators currently read the request, specialists interpret technical clauses, a proposal manager owns the matrix, and a red-team reviewer challenges evidence and claims.

In the first controlled case, extraction reduces manual copying but places ambiguous requirements into a generic category. Specialists spend additional time reopening source pages, while the proposal manager assumes the matrix has already been checked. The map identifies a missing decision right: nobody is explicitly responsible for resolving ambiguous classification before drafting begins.

The future-state design keeps extraction with the assistant, assigns clause interpretation to a named specialist, and requires the proposal manager to confirm source links before release. The organizational development leader facilitates the redesign and schedules a retrospective; the proposal lead owns the operating decision.

A second synthetic case shows fewer unresolved handoffs, but no production-volume or customer evidence exists. The decision is to continue a controlled pilot, not declare a successful transformation. The example is synthetic and is not evidence of Two Prune customer impact.

Verify the work system, not the diagram

Observe representative cases and compare actual roles, queues, overrides, workarounds, review effort, and escalation with the map. Ask affected workers where judgment or invisible labor moved. Check whether new responsibilities have authority, time, training, accessible tools, and a route to stop unsafe release.

Reopen the map after changes to workflow, model, data, policy, volume, team, or consequence. Two Prune can provide task-bounded evidence about participant behavior in controlled simulations for human interpretation. It does not redesign customer organizations, prove transformation outcomes, or make employment decisions.