As companies race to deploy generative AI across their organizations, many executives are betting on a transformative promise: that GenAI will allow employees from one function to seamlessly take on work traditionally performed by specialists in another, and perform it at a level that matches the specialists themselves.
The logic is appealing: If a marketing professional can suddenly perform data analysis, or if an engineer can produce compelling marketing content, companies could achieve unprecedented workforce flexibility and efficiency. But new research suggests that this vision has critical limits that executives must understand before reorganizing their talent strategies.
In a field experiment conducted at IG, a leading U.K. fintech company, with analysis by researchers from Harvard Business School, Stanford University, and the Stanford Digital Economy Lab, participants examined whether GenAI could enable professionals from different occupational backgrounds to perform tasks at the same level as specialists. The experiment recruited employees from three distinct groups: web analysts who regularly write content for the company’s website (the “insiders”); marketing specialists who work in related functions but don’t typically write web articles (“adjacent outsiders”); and technology specialists—data scientists and software developers—whose work is entirely unrelated to content creation (“distant outsiders”).
All participants were asked to complete two sequential tasks: first, conceptualizing an article (outlining structure, keywords, and key points), and then executing the full article. Some participants had access to IG’s bespoke GenAI tools; others did not.
The results reveal what we call the “GenAI wall effect”—a threshold beyond which GenAI can no longer meaningfully bridge the expertise gap between specialists and non-specialists. Understanding where this wall emerges is essential for any company seeking to leverage AI for workforce transformation.
When GenAI breaks down occupational boundaries—and when it doesn’t
The experiment yielded a striking pattern. For the conceptualization task, creating the article’s outline and structure, GenAI effectively eliminated performance differences across all three groups. Without AI assistance, web analysts significantly outperformed both marketing and technology specialists. But when equipped with GenAI, marketing specialists and technology specialists produced conceptualizations statistically indistinguishable from those of the experts. GenAI acted as a powerful equalizer for this type of abstract, structured work.
This pattern aligns with what we have called in a different experiment AI’s “jagged technological frontier,” the uneven boundary between tasks where AI performs well with minimal human guidance and those where it cannot.
However, when it came to execution, a task that falls outside that frontier, the results diverged dramatically. Marketing specialists, when aided by GenAI, were able to produce articles of comparable quality to the web analysts. But technology specialists could not. Even with full access to the same AI tools, they consistently underperformed.
This is the GenAI wall in action: The technology could bridge the gap between “adjacent” outsiders and insiders but hit a hard limit when the knowledge distance became too great.
Why knowledge distance determines GenAI’s effectiveness
Post-experiment interviews revealed why this wall emerged. The three groups approached the tasks with fundamentally different mental models rooted in their professional backgrounds. Web analysts and marketing specialists shared a common vocabulary around customer engagement, conversion optimization, and audience targeting—they understood intuitively what makes marketing content effective. Technology specialists, meanwhile, approached the writing task as they would technical documentation: prioritizing brevity, clarity, and directness.
This difference proved consequential when editing GenAI outputs. Marketing specialists used the AI-generated content as a foundation they could evaluate and refine because they possessed the foundational knowledge to judge quality. Technology specialists, lacking this domain expertise, often made edits that inadvertently degraded the content—removing “marketing spin” they didn’t recognize as valuable, shortening articles below optimal length for SEO, and eliminating calls-to-action they viewed as unnecessary.
As one data scientist candidly admitted: “GenAI suggested some catchy hooks… Actually, I didn’t fully understand what it was doing because I never wrote an article like that. I added random stuff to make it more ‘marketing.'” Another explained removing large portions of AI-generated content because he “prefer[red] articles that are clear and direct”—precisely the opposite of what effective marketing content requires.
In essence, professionals with domain expertise knew the destination and used GenAI to help chart the route. Those without it had to trust the AI for both navigation and the final destination—and that’s where things went wrong.
The critical distinction: conceptualization versus execution
Why did GenAI succeed in closing the gap for conceptualization but not execution? The answer lies in the fundamental nature of these tasks. Conceptualization is an act of structured abstraction—listing features, identifying keywords, outlining structure. It follows a template, and GenAI excels at providing reasonable suggestions that even novices can evaluate.
Execution, by contrast, is an act of embodiment—transforming abstract ideas into tangible, polished prose. This requires not just generating content but making countless micro-judgments about tone, emphasis, audience appeal, and strategic intent. These judgments depend on tacit knowledge that comes from domain experience. Without that foundation, users cannot effectively evaluate, refine, or improve AI-generated output.
This finding challenges the popular notion of an “ideation-execution gap” that assumes AI-generated ideas simply fail in implementation. Our research suggests the opposite: The bottleneck isn’t the idea’s quality—it’s the implementer’s knowledge distance from the domain.
What this means for executives
These findings carry significant implications for how companies should approach GenAI-enabled workforce transformation.
First, be realistic about cross-functional mobility. GenAI can facilitate redeployment between adjacent functions, where employees share foundational knowledge, but not between distant ones. A marketing coordinator may successfully transition to content creation with AI assistance; a software developer likely cannot, at least not without substantial retraining. Companies should map their functions by knowledge distance before assuming AI will enable frictionless workforce flexibility.
Second, recognize GenAI’s power to accelerate learning curves. While the wall limits cross-functional substitution for distant roles, GenAI can dramatically shorten—or even eliminate—the ramp-up time for adjacent tasks. This makes it a powerful tool for expanding the scope of existing roles, even when it cannot replace specialists outright.
Third, distinguish between task types. GenAI democratizes conceptualization and ideation much more effectively than it does execution. Organizations can leverage this by separating these phases: Use AI-augmented diverse teams for ideation, but route execution to those with relevant domain expertise.
Fourth, invest in foundational knowledge. The GenAI wall isn’t about technical AI skills—it’s about domain expertise. The technology specialists in our study were highly capable with AI tools yet still underperformed because they lacked marketing fundamentals. Training programs should prioritize building domain knowledge, not just AI proficiency.
Fifth, rethink what defines expertise. Our research suggests a shift in what makes someone valuable: from hands-on repetitive practice toward broader foundational knowledge that enables effective human-AI collaboration. Hiring and development strategies should evolve accordingly.
Finally, resist the allure of total workforce fungibility. The promise that GenAI will eliminate all specialization is seductive but unrealistic. The technology creates new possibilities for horizontal expertise transfer—but only across certain boundaries. Deep domain knowledge remains essential for complex execution tasks, and likely will for some time.
***
The GenAI wall effect reveals that artificial intelligence, for all its power, cannot fully substitute for human expertise—at least not yet. The technology excels at providing information, generating options, and supporting structured tasks. But converting those inputs into high-quality outputs still requires the kind of foundational knowledge that humans acquire through education and experience in specific domains.
For executives, the strategic imperative is clear: Understand where your organization’s GenAI walls exist. Map the knowledge distances between your functions. Identify which boundaries AI can dissolve and which it cannot. Then design your talent, training, and organizational strategies accordingly. And because AI capabilities are improving rapidly, the wall is not fixed, it will shift over time, making regular reassessment essential. Companies that get this right and keep reassessing will achieve genuine competitive advantage. Those who assume GenAI eliminates the need for expertise will hit the wall and wonder why their workforce transformation stalled.
Francois Candelon is a partner at private equity firm Seven2 and executive fellow at the HBS AI Institute (formerly known as the D^3 Institute). Read other Fortune columns by François Candelon.
Iavor Bojinov is the James Dinan and Elizabeth Miller Associate Professor of Business Administration at the Harvard Business School.
