Scaling AI starts with redesigning work, not adding tools
There is a narrative circulating in many organisations at the moment and on the surface it sounds positive. People feel more productive because they are experimenting with AI assistants, drafting content more quickly, analysing information faster and automating tasks previously done manually. In conversation, individuals often speak with genuine enthusiasm and report experiencing meaningful benefits from these tools.
But when we step back and look at the organisation as a whole, the story becomes less clear. Although adoption is now widespread, with most businesses using AI in at least one area, McKinsey’s recent research shows that enterprise-level value remains limited, with only modest impact on productivity, profit or operating performance. That gap is prompting concern, and leaders are beginning to ask an uncomfortable but necessary question: if everyone is working faster, where is the value going?
Where productivity gets stuck
Part of the issue is that we have been talking about AI as if it is one single capability. In reality, the term covers very different applications that demand different skills, governance and organisational support. There is the long-established world of machine learning, built on data and mathematical modelling, and agentic or autonomous AI, which executes steps within workflows and makes constrained decisions. Alongside these are personal and team productivity tools, now widely used with little formal training. All three sit under the AI banner, but they are not interchangeable and treating them as such fuels confusion and unrealistic expectations.
This is where the paradox begins to take shape. A recent MIT Media Lab study found that despite a surge in AI adoption, 95% of organisations have yet to see measurable returns. The enthusiasm is real, but the value is not flowing into business performance because organisations prioritise individual productivity, which is easier to trial. People work faster and feel the benefits directly, yet individual efficiency does not translate into systemic improvement.
In many cases, this approach accelerates existing behaviours. If a report once took two days and now takes an hour, the instinct is to produce more reports. What often goes unchallenged is whether the work itself is needed, resulting in more output that creates noise rather than value.
Formal processes rarely reflect how work actually gets done. In the gaps left by systems and governance, people create shortcuts, spreadsheets and informal agreements. These shadow processes represent the true operational reality, and when AI is layered onto them, it amplifies fragmentation rather than correcting it.
This is why AI success is less about technology adoption and more about organisational architecture. To make progress, businesses need to understand how work genuinely flows, not how it appears on a slide. That means examining decision points, handovers and rework, and accepting that some processes should not be automated at all.
As organisations begin this work, new roles are emerging at the boundary between people and AI. Workflow interpreters and human-in-the-loop supervisors help translate intent, guide use and maintain oversight, playing a critical role in moving AI beyond experimentation and into everyday operations.
The leadership mandate
Leadership plays a decisive role in guiding this transition. It may feel natural to approach AI as another technology rollout with milestones and a clear endpoint, but AI does not behave like traditional software. It evolves and exposes weaknesses in culture, clarity and process. For leaders, this requires a shift from implementation to system design, and the creation of environments where experimentation is expected rather than feared.
As AI becomes more embedded, employees are already questioning where their value sits. It’s a pattern echoed in EY’s recent research, which found that despite widespread day-to-day use of AI tools, only 28% of organisations have redesigned roles, workflows or operating models in ways that enable employees to generate real business value from them. Naturally, workers are increasingly led to wonder whether their contribution is measured by the volume of output or whether their judgement, creativity and contextual understanding hold greater weight.
These questions reflect a cultural rather than technical shift. Leaders must communicate a vision where AI augments talent rather than replaces it. Productivity metrics alone cannot resolve concerns about identity or relevance. Employees need clarity on how AI fits into meaningful work, and reassurance that skills such as critical thinking, ethical awareness and human connection remain central.
This is where HR and organisational design leaders can have the greatest impact. They understand how roles evolve, how culture is shaped, and how systems and behaviours interact. Their task is to move organisations beyond a narrow focus on efficiency, toward a more sophisticated understanding of what AI enables. The goal is not simply faster work, but better work that delivers deeper insight rather than reduced effort.
To achieve this, businesses must redesign themselves so individual experimentation translates into collective capability. That requires clear workflows, shared knowledge and frameworks that allow innovation while maintaining governance. Employees also need psychological safety and the confidence to question AI outputs, knowing when to trust, challenge or override them.
Tools may accelerate work, but leadership, structure and culture determine whether that acceleration creates clarity or chaos.
—
