The Moving Frontier — On AI, business, and a changing world. By Dhruv Dhingra.

You Can’t Prompt Your Way to Organizational Change

By Dhruv Dhingra


THE MOVING FRONTIER · NOVEMBER 2025 RETROSPECTIVE

A tool is not a transformation

A new AI tool can produce a striking demonstration: a faster document, a convincing answer, a few minutes saved. Yet organizations that celebrate those wins can find that the underlying way of working hardly changes.

The missing ingredient is seldom another prompt. It is an operating model. When an intelligent system begins doing pieces of the work, the organization must decide what remains human, who owns the result, and how quality will be judged.

Those are management and product-design questions as much as technical ones.

Work is more than tasks

Take a seemingly simple process such as preparing a quarterly business review. The visible task is assembling slides. The less visible work involves choosing which metrics matter, checking disputed numbers, reconciling different narratives, and earning agreement about next steps.

Automating slide production may save time. It does not by itself improve the decision. If the organization continues to reward beautiful decks over useful conversations, the technology will accelerate the old behavior.

The right unit of change is the end-to-end decision process, not the easily demonstrable task.

Redraw the accountability map

I would start by naming the accountable human for an outcome, then separating the stages of work. What can be delegated safely? What demands expert review? What evidence must be retained? Who investigates an exception? Who decides whether an automation should be expanded or paused?

Without answers, ambiguity accumulates. One team thinks the system is an assistant, another treats it as an author, and a third assumes that somebody else is reviewing its work.

The organization needs clear ownership of both quality and improvement. An AI system that participates in a workflow should have monitoring, incident response, and a path for resolving recurring failure modes.

Change the incentives

Adoption fails when a new workflow creates costs for the people asked to use it. If a practitioner must spend more time checking outputs than they save producing them, usage will fade. If managers continue evaluating employees by the volume of manual artifacts, automation will feel threatening rather than helpful.

Successful change includes training, new review habits, redesigned measures of productivity, and permission to stop activities that no longer create value. It should also acknowledge that some tasks are valuable precisely because they help junior staff develop judgment.

That last point matters. A team can automate the visible work and inadvertently remove the apprenticeship that creates its future experts. Organizations should make the learning pathway explicit.

The real question

The question is not ‘How many people tried the AI tool?’ It is ‘Which decision or workflow has become materially better, and can we prove it?’

Tools can be adopted individually. Operating models must be designed collectively. No prompt can do that work for a leadership team.

Editorial note: This is a retrospective essay written in November 2026. November 2025 describes the period explored, not its original publication date.


← All essays