Gen AI Boosts Productivity, But Can't Turn Novices Into Experts

Workplace Technology · AI & the Future of Work · March 2026 · 8 min read

There is a seductive promise baked into every generative AI demo: that anyone regardless of background or experience can produce work that rivals a seasoned professional. Write like a journalist. Code like an engineer. Strategise like a consultant. Just ask the machine.

A new study from Harvard Business School offers a more nuanced verdict. Yes, AI dramatically expands what workers can attempt. But whether their output is any good depends heavily on how close their existing skills are to the task at hand.

"AI makes you feel like you can do anything. But can you do the task as well as people whose job it is?" — Professor Iavor Bojinov, Harvard Business School

What the research found

Researchers studied 78 workers at IG Group, a global derivatives trading firm, and divided them into three groups: web analysts (insiders), marketing specialists (adjacent), and technology specialists developers and data scientists (distant). All three groups were given the same AI tools and asked to write investing articles for IG's website.

The results landed on a clear fault line. Marketing specialists, whose skills overlapped significantly with the analysts, produced articles that scored nearly on par with the professionals. Technology specialists, despite being highly skilled in their own domain, lagged behind by about 13% on clarity and writing competence.

The "knowledge distance" problem

The researchers introduced the concept of knowledge distance to explain the gap. Think of it like instruments: a flautist picking up an oboe faces a far smaller learning curve than if they were handed a violin. The same principle applies to knowledge work adjacent skills transfer with AI's help, but distant ones hit a wall.

A marketing specialist already understands content framing, audience tone, and narrative structure. When AI fills in the gaps, they can bridge to analyst-quality output. A data scientist, however, lacks the lived editorial instincts that writing requires. The AI provides the map, but they don't yet know how to read the terrain.

"The folks who are too far away from the domain experts lack either sufficient understanding of the necessary information or the skills to use it effectively." — Professor Edward McFowland III, Harvard Business School

Where AI levels the playing field and where it doesn't

Where AI helps: Brainstorming, outlining, and structuring ideas. All three groups performed equally well during the conceptualization phase averaging scores between 4.05 and 4.18 out of 5. Structuring information is abstract and codifiable it maps naturally to how technical minds think, and AI can scaffold this process effectively for almost anyone.

Where AI hits a wall: Execution the craft of writing, applying nuance, and making judgment calls in context. Writing is context-bound and creative. Without the intuitions built from years of doing the actual job, novices can't fully interpret or act on the AI's guidance. This is where lived experience still separates experts from outsiders.

The productivity gains are real and remarkable

Even where quality diverged, the time savings were universal.

• •Conceptualization time fell by nearly two-thirds from 63 minutes to 23 minutes on average.
• •Writing time dropped by almost three-quarters from 87 minutes to just 22.

Every worker, regardless of background, got significantly faster.

This matters. Even if a data scientist's articles are 13% weaker than a web analyst's, the speed gains may well justify deploying AI-assisted workflows in contexts where volume matters more than perfection, or where the task is conceptual rather than editorial.

What this means for organisations

The findings carry practical implications for how companies think about hiring, training, and job design as AI becomes embedded in everyday workflows.

Retraining costs shrink for adjacent moves. A data scientist transitioning to a marketing analytics role faces a much shorter learning curve with AI than they would have even five years ago but their path to becoming a content strategist remains steep. Organisations that map their roles carefully will deploy AI far more effectively than those treating it as a universal upskiller.

Flat organisations become more viable. When AI can compress learning curves for tasks like SEO writing or structured analysis, some middle layers of specialist roles may become thinner. Generalists with adjacent knowledge become more valuable because AI makes their adjacent skills do more work.

Ideation becomes a collaborative strength. For brainstorming, strategy sessions, and early-stage project framing, AI is a genuine equaliser. Companies should actively build this into their workflows and stop waiting for the technology to fix execution problems it structurally cannot solve.

Key takeaways

• •AI can help workers with adjacent skills match expert output quality but not workers with distant skills.
• •Conceptualisation tasks (structuring, outlining, framing) are where AI helps everyone most equally.
• •Execution quality still depends on lived experience and domain expertise.
• •Time savings are substantial and universal regardless of skill proximity.
• •Organisations should rethink training, role design, and hiring with these limits explicitly in mind.