Three myths that undermine AI success

The real AI story in most organizations isn’t about algorithms; it’s about habits. New tools arrive with impressive demonstrations and confident promises, yet the day-to-day routines that decide what gets attention, who can take a risk, and what counts as a “good job” tend to remain the same. Leaders set up special units, roll out training, or look for quick savings, only to find that the old culture quietly resets the terms. When that happens, early gains fade, adoption stalls, and cynicism grows.

This article draws on our forthcoming book to look at three recurring myths that help prop up existing cultures and prevent the deep transformations that are needed to support successful AI implementations. Transforming a business to make the most of AI means moving past these comfortable stories and changing the conditions under which the whole organization works.

Myth 1: ‘Innovation Units Will Save Us’

After five years of operation, the U.K.’s Government Digital Service (GDS) seemed untouchable. Created in 2011, the GDS revolutionized Britain’s digital services. With the goal of reenvisioning “government as a platform,” it consolidated hundreds of websites into a single, easy-to-use portal, cut waste by forcing departments to unify their platforms, and showed that, with the right attitude, even government agencies could move with the speed of a startup. In 2016, the U.K.’s digital services were ranked the best in the world. Yet by 2020, the GDS had disappeared as a force within the U.K. government.

This pattern repeats regularly across corporate innovation labs: create an elite unit, give it special rules, celebrate early wins, watch it die. An innovation unit can deliver extraordinary results so long as it has senior leadership protection, free-flowing resources, and an internal culture that attracts exceptional talent. But the model also contains the seeds of its own demise. The outsider status that enables breakthrough innovation makes large-scale sustainability nearly impossible. When executive sponsors move on, the shield drops, and organizational antibodies start reasserting cultural norms.

This predictable lifecycle applies to AI-focused teams as much as those driving any other type of technological change. Leadership transitions are inevitable. New executives question special rules. The innovation unit that draws its power from being outside the system gets pulled back in again, and the flow of novel ideas slows to a trickle.

The lesson to take from this isn’t that we should abandon innovation units—it’s that we should use them strategically and follow up on the gains they make. Innovation units should be seen as catalysts, not permanent solutions. While these teams are forging ahead with quick wins and proving new approaches, organizations also need to transform their broader culture in parallel. The goal shouldn’t be protecting the innovation unit indefinitely but aligning organizational culture with the innovative approaches it pioneers. If innovation units are sparks, culture is the oxygen. You need both—at the same time—or the flame dies.

Myth 2: “Our People Just Need Training”

Companies spend millions teaching employees to use AI tools, then wonder why transformation never happens. The reason is that the underlying problem isn’t just about skills—it’s about the imagination needed to use them effectively. You can train your workforce to operate the new technology, but you can’t train them to be excited about it or to care where it will take the business. That requires change at the cultural level.

When it comes to AI, the real gap is conceptual, not technical. Employees need to shift from seeing AI as a better calculator to understanding the role it can play as a thought partner. This requires more than tutorials. It means showcasing how AI can transform workflows and then rewarding its creative use. Show a sales team how AI can predict client needs before calls, not just transcribe them afterward. Demonstrate how legal teams can shift from document review to strategic counseling.

When organizations tell employees to “use the tools” but don’t change the social norms around using them, people can be punished for doing exactly what leadership asked. A recent experiment with 1,026 software engineers found that when reviewers believed code was produced with AI assistance, they rated the author’s competence lower by about 9% even though the work was identical. Even more concerning was that the penalty was larger for women and older engineers, groups who tended to be treated negatively in assessments already. In a companion survey of 919 engineers, many reported hesitating to use AI for fear that adoption would be read as a lack of skill—illustrating why access and training don’t translate into uptake when the culture signals that visible AI use will harm credibility.

Myth 3: “AI Makes It Easy to Slim Down the Workforce”

There’s a seductive promise being sold to companies right now. The way to realize AI’s value is simply to replace as many workers as you can. Fire half your staff, pocket the savings, let machines handle the rest. Simple arithmetic for simple minds.

The messy truth is that AI can and will replace many human jobs, but it won’t do it cleanly and it won’t do it easily. In most cases, the idea that you can simply swap out the human component and replace it with a machine just doesn’t work. Humans work together as parts of multilayered social structures that have evolved as ecosystems. It’s often the case that if you change one part, there will be major consequences for another. If we rush into automation too quickly, we risk pulling away the pillars that hold the whole structure up.

Think about the tedious hours that junior analysts spend cleaning data, checking figures, and building models from scratch. Or the work a newly appointed manager will do overseeing performance and filling in paperwork. We call it grunt work, but it’s actually how humans develop the skills they will need in more senior roles. Take away the entry-level jobs and you lose the career path that delivers the highly skilled senior leaders you need. Allow AI-powered “deskilling” to take place and you lose the human judgment and oversight that institutions rely on.

Klarna’s trajectory shows both sides of this equation. In early 2024, its AI assistant handled two-thirds of customer chats, delivering resolution times under two minutes and a 25% drop in repeat inquiries. By 2025, Klarna’s leadership was publicly acknowledging the limits of an AI-only approach and began reopening human roles and emphasizing the customer experience alongside automation.

The real question isn’t how many people you can eliminate. For effective AI implementation, you need to understand that humans make essential contributions that don’t appear in their job descriptions.

The Culture Transformation Playbook: Fixing the Myths

Culture change depends on habits, incentives, and expectations, not just adding new tools. The playbook that follows presents concrete steps that leaders can take now to avoid the pitfalls many companies are running into.

• Run Parallel Transformations (Fixes Myth 1). The innovation unit delivers quick wins while a separate initiative transforms broader culture. These must happen simultaneously, not sequentially. Use the innovation unit’s protected status and early victories to create organizational belief in change but invest equally in preparing the mainline culture for what’s coming. Without parallel tracks, the innovation unit becomes an isolated island of excellence that will eventually be washed away.
• Transform the Middle Layer (Fixes Myth 2). Middle managers are the real gatekeepers of culture change. Stop wasting energy trying to convert skeptics. Instead, identify the curious and give them authority to experiment, budget to fail, and cover from meeting traditional metrics. Try giving selected managers a micro-charter to implement change in their team, along with a weekly “show-the-work” session (what AI was used, what was accepted or overruled, and why) to share what they’ve learned with peers.
• Build Alternative Learning Paths (Fixes Myth 3). If AI eliminates the experiences that build judgment, you must consciously re-create them. High-fidelity simulations, rotation programs, and “human days” working without AI become existential necessities. Explicitly preserve activities that develop pattern recognition and business instinct. The investment might seem wasteful until you realize the alternative is a workforce that can operate tools but can’t respond when something breaks.

The Choice

Culture transformation is harder than technology implementation. It’s messier, slower, and impossible to fully control. Most companies will choose the easy path: buy the AI, train on the tools, create an innovation lab, and hope for the best.

The few who choose the hard path—parallel transformation, cultural evolution, preserved learning experiences—will gain powerful competitive advantages. They’ll have workforces that don’t just use AI but think with it, cultures that don’t just tolerate change but expect it, and organizations that don’t just survive disruption but drive it.

View the full article