Why Agentic Business Transformation Starts with Organisational Design

Most enterprises struggling with agentic AI have the same problem. They deployed the technology on top of everything that already existed.

Same org chart. Same approval chains. Same performance metrics designed to count what a person does in a day. AI agents sit inside structures they were never built for — and then leadership wonders why the ROI isn't materialising.

85% of organisations say they want to be agentic within the next three years. But 76% say their current operations and infrastructure can't support that change. The gap between ambition and execution is not a technology problem. It is an organisational design problem. This is the gap MIT Technology Review explored in a recent article, published last week in partnership with Ema.

The Sticky Tape Problem

PwC's global CTO for workforce consulting and Chief AI Officer, Prasun Shan, has a phrase for what most enterprises are doing. He calls it sticky tape.

"They're embedding AI employees into what is a human operating model. This is like adding sticky tape to parts of an operating model that is breaking." — Prasun Shah, Global CTO for Workforce Consulting & Chief AI Officer, PwC UK

The problem with sticky tape is structural. If human bandwidth is still the binding constraint on approvals, escalations, and handoffs, AI assistance compresses cycle time at the margins. It doesn't change what the organisation can do at scale. AI assistance is not AI transformation.

The value in agentic AI lies in its capacity to execute entire workflows with limited human input — to coordinate complex tasks, make independent decisions, and iterate performance. It's already estimated that AI agents could accelerate business processes by 30–50% when deployed at scale. Getting to that value requires redesigning how the organisation works, not just adding AI to how it currently works.

What Agentic Business Transformation Actually Means

Last year, Ema, in partnership with HFS Research, coined the Agentic Business Transformation (ABT)—because none of the existing vocabulary captured what actually needs to change to succeed with Agentic AI. (You can explore the full ABT framework at agenticbusinesstransformation.ai.)

"Digital transformation was about moving from paper to software. AI transformation was about adding AI to existing processes. Copilot is about AI assisting humans. But ABT is something categorically different: it's the integration of AI agents into the fabric of the organisation." — Surojit Chatterjee, CEO and Founder, Ema

ABT is built on a five-level Agentic Maturity Model — from basic task agents to fully autonomous AI workforces — and it requires redesigning three things simultaneously.

Pillar One: The Tech Stack

The existing technology stack was designed for human-operated, application-centric workflows. AI agents don't fit that model. Their value isn't as another layer on top of the stack — it's as connective tissue that moves between layers, coordinating high-level tasks and retrieving data across multiple systems simultaneously.

Organisations that make this architectural shift don't wait months for a software vendor to build a new feature. They configure an AI employee in natural language and connect it to the systems it needs. The time from business requirement to production workflow drops from months to days.

Pillar Two: The Workforce

When AI agents can execute, coordinate, and optimise without managerial coordination, hierarchies blur. Managers take on new responsibilities: managing trust, explainability, and psychological safety in hybrid teams. McKinsey estimates that three-quarters of current jobs will require redesign, upskilling, or redeployment by 2030.

Pillar Three: The Metrics

Activity metrics become misleading the moment AI employees enter the picture. An AI employee handles a thousand customer interactions in the time a human handles ten. Measuring interactions tells you the AI is fast. It tells you nothing about whether any of those interactions drove satisfaction, retention, or revenue.

When one of Ema's enterprise customers switched from metrics like cost per query and AI accuracy to outcomes like the percentage of contracts reviewed without human escalation, their measured ROI from agentic AI tripled within two quarters.

The Groundwork for Systems-Level Change

The organisations pulling ahead aren't waiting for certainty. They are redesigning for what work and the organization should look like. If you're responsible for running agentic AI in your enterprise, here's how you can learn how to lead through the shift:

• The ABT Manifesto for the principles behind this shift
• The ABT resource library, with 5 levels of agentic maturity and other key frameworks
• Rethink Organizational Design for the Age of Agentic AI on MIT Technology Review Insights