Why AI Transformation Gets Stuck in Old Ways of Working
When I made this image (using AI, of course), my initial reaction was to laugh. Then I thought to myself, why is this funny? We have a rocket labeled AI, tied to an anchor labeled “How we’ve always done it,” and apparently that is enough to summarize a fairly expensive business problem. I also spent time asking AI to make the lettering thicker, then thinner, so I recognize that my own contribution to humanity’s technological progress requires some perspective.
What stayed with me was how reasonable the anchor can sound. Nobody walks into a meeting and announces that they would like to prevent progress. They ask whether the new approach can follow the existing approval process, fit within the current organizational structure, and leave everyone’s responsibilities intact. Individually, those can be sensible questions. Collectively, they can turn “transform the business” into “please make our current business slightly faster without disturbing anything we have become attached to.”
I love technology, which probably isn’t a shocking revelation to anyone who follows me. Give me something that connects, automates, analyzes, or does something unnecessarily impressive, and you have my attention. That enthusiasm is useful, although it also makes it easy to spend more time discussing what a tool can do than what the organization will actually allow to happen. A demonstration rarely includes the part where three departments disagree over whose budget should pay for the improvement.
Consider an AI recommendation that takes ten seconds to generate and two weeks to approve. The analysis may be more useful, and the time saved may be real, but the customer still waits for a decision. Somewhere, a project dashboard can show a large productivity improvement while the customer experiences approximately none of it. I keep coming back to that disconnect because making one task faster can feel like progress even when the overall experience barely changes.
The 2026 research gives that disconnect some numbers. In McKinsey’s August 2026 State of AI survey, 80 percent of respondents said AI had improved their individual productivity, yet 37 percent reported a positive contribution to their organization’s earnings before interest and taxes. Those measure different things, and the gap doesn’t mean everyone else got no value. It does make “we saved people time” a less satisfying stopping point. I want to know where that time went, what changed for customers, and whether we improved the business or simply gave everyone the capacity to attend another meeting.
And this is where the human part gets complicated. Microsoft’s 2026 Work Trend Index, which surveyed 20,000 knowledge workers using AI across ten countries, found that 45 percent felt safer concentrating on existing goals than redesigning their work with AI. That makes sense when someone’s expertise, reputation, and promotion are connected to a process they spent years improving. If I ask them to help eliminate it while measuring their importance by how many decisions pass through them, I have given them a fairly confusing assignment. Calling their response “resistance to change” lets me avoid examining the incentives I helped create.
And yes, this applies to leaders too, including me. It is much easier to enthusiastically support empowerment in a presentation than to accept a good decision that happened without your involvement. I think we have to get comfortable with the possibility that something we approved, sponsored, or proudly built can eventually become unnecessary. Its usefulness ending does not mean it was a mistake, although our attachment to being right can make those two things feel remarkably similar.
I would start with one important workflow and follow it all the way through, especially across departmental boundaries. Where does work wait, who can move it forward, and what happens when the recommended action benefits the company while making one team’s numbers look worse? Those questions belong in the same conversation as the technology. So does a practical discussion with the people doing the work about what they will own next, what training they need, and how success will be measured when the old activity disappears.
Of course, some anchors are doing an important job, particularly in physical operations where a bad decision can hurt someone. The point is to understand which controls protect an outcome and which survive because nobody remembers who is allowed to remove them. Looking at the image now, I find myself wondering how often we ask technology to overcome something we still have the authority to change ourselves. Before buying a more powerful rocket, I would quite like us to have that conversation.
References:
Microsoft. (2026, May 5). Agents, human agency, and the opportunity for every organization. https://www.microsoft.com/en-us/worklab/work-trend-index/agents-human-agency-and-the-opportunity-for-every-organization
Tinkoff, D., Van der Veken, L., Chui, M., & Balakrishnan, T. (2026, August 25). The state of AI in 2026: On the road to ROI. McKinsey & Company. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
