Every time a wave of automation hits, the same fear surges: the machines are coming for our jobs.
McKinsey’s latest AI report lands with a sobering qualifier. Productivity gains are real, but conditional. They appear only when organisations redesign their workflows around AI, not when they bolt AI onto what already exists. Most companies are still bolting.
The Pattern Across the Boardroom
I have seen this pattern across boardrooms in Hong Kong, Singapore, and the mainland. The organisations struggling most with AI ROI are treating it as a technology upgrade. The ones moving fastest are treating it as an organisation design challenge. Same AI. Very different results.
The distinction sounds clean in a research report. In practice it is harder to see from the inside, because the organisations in bolt-on mode are not obviously doing it wrong. They are subscribing to the right tools, running the pilot programmes, attending the conferences, and reporting positively on AI adoption in their board packs. The metrics look fine until someone asks the honest question: has anything about how this organisation works actually changed?
Usually, the answer is not much.
Why Bolt-On Fails
The reason bolt-on AI implementations underperform is structural. AI tools are, at their best, capable of taking over the cognitive labour that sits at the edges of workflows: the research, the first draft, the data synthesis, the pattern recognition across large information sets. But those tasks are rarely the bottleneck in a high-performing knowledge work organisation. The bottleneck is usually the decision-making layer, and AI cannot own the decisions.
When you add AI capability to an organisation that has not redesigned how decisions get made and how work flows between people and systems, what you get is faster execution of the existing structure. Faster is good. Faster-but-unchanged is not transformation.
The organisations that are getting real returns are doing something harder. They are asking which roles need to exist, which workflows need to be rebuilt from the ground up, and what human judgment is genuinely irreplaceable at each stage of the value chain. Those are uncomfortable questions, because the answers often challenge structures that have existed for a long time and that people have built careers around.
The Historical Pattern Worth Remembering
Here is what gets lost in the displacement debate: every major productivity technology in history, from the printing press to the industrial loom to the personal computer, eventually created more work than it eliminated. Not always immediately. Not always for the same people. Not always in the same geographies. But the net direction has been consistent across every major technological transition.
AI will follow the same arc, with one important caveat: the organisations and leaders who shape that arc proactively will have significantly better outcomes than those who respond reactively. The technology does not determine what happens. The decisions made about how to deploy it, and who gets to be part of that process, determine what happens.
What Redesigning Workflows Actually Means
When McKinsey talks about redesigning workflows, the phrase can sound like consultant shorthand for a process reengineering project. In practice, the organisations doing it well are asking a more fundamental set of questions.
Which parts of this role currently require human judgment, and which parts are essentially mechanical transformation of information? If a workflow is rebuilt around AI doing the mechanical parts, what does the human part of that role become? Do the people in those roles have the support and development to grow into what the job is becoming?
These are not technology questions. They are people and culture questions. They require the kind of leadership honesty that is harder to deploy in a quarterly planning cycle than a pilot programme.
The Leadership Imperative
To me, this is the leadership imperative of the current moment: not to manage AI as a tool, but to architect the human organisation that works alongside it.
That means redesigning team structures from the ground up. Rethinking job roles, not just which tasks get automated, but what human judgment and creativity should fill the time that AI frees up. It means having honest conversations with people about how their work is changing before those changes happen, not after.
The hardest part is not the technology. It is having the honesty and courage to rebuild how work actually gets done, with people at the centre of that design.
The gains will follow. They always do, for the leaders willing to start with people first.