The Implementation Gap in AI Learning Concerns

Lynda Gratton asks exactly the right questions in her recent HBR piece on AI and workplace learning. Her provocation that “accelerated learning is not the same as development” cuts to something we’ve been wrestling with for years: the difference between knowing and capability, between exposure and transformation.

We agree completely with her core diagnosis. When AI handles difficult conversations, drafts strategy memos, and generates the first 20 ideas, organizations aren’t just gaining efficiency—they’re systematically dismantling the struggle that builds expertise. Gratton’s parallel to pandemic technology is particularly sharp. Just as Zoom expanded meeting volume rather than thinking quality, AI expands content volume while crowding out the reflection that produces insight.

What strikes us most is her observation about empathy development. “Empathy grows through practice,” she writes—”reading subtle cues, managing conflict, engaging in difficult conversations.” This aligns precisely with what implementation science tells us about capability formation. Amy Edmondson’s research on psychological safety, Deci and Ryan’s work on intrinsic motivation, decades of expertise research—all point to the same pattern: capability emerges from repeated exposure to challenge, not from having challenge removed.

Where We’d Push Further

Gratton offers “sense-making conversations” as her solution—provocations that help leadership teams ask difficult questions. We think she’s right that questions matter more than answers in uncertain territory. But questions alone leave organizations without protocols for action.

Thanks for reading Lead from the Front! This post is public; please feel free to share the knowledge.

This is the gap we’d name: the implementation gap. Leaders may leave Gratton’s workshops understanding intellectually that AI threatens developmental pathways. They may genuinely want to “preserve space for human choice.” But wanting something doesn’t create the organizational capability to do it—especially when AI adoption pressure is immediate and developmental timelines are measured in years.

Our research suggests organizations need specific protocols, not just provocations.

• First, conduct a theory-in-use audit on AI deployment. Organizations say they value human development, but what do their AI adoption decisions reveal they actually believe? If every workflow redesign prioritizes efficiency without explicitly preserving development, the organization’s theory-in-use is that development doesn’t matter—despite the stated values. Audit the gap. Make it visible.
• Second, build developmental friction into AI-assisted workflows deliberately. This means identifying which struggles produce capability and structurally protecting them. Not all friction is developmental—some is just waste. But the junior analyst wrestling with a dataset, the manager navigating a difficult conversation, the strategist facing a blank page—these struggles build the neural pathways that enable judgment under pressure. Design systems that use AI as scaffolding rather than replacement.
• Third, measure what matters over timelines that matter. Behavior change research consistently shows 66+ days minimum for habit formation, 6+ months for complex skill development, and 18-24 months for capability that persists under stress. Organizations measuring AI’s impact on development need leading indicators tracked over these timelines—not quarterly satisfaction surveys. Track how often employees face genuine decision points. Track whether psychological safety increases or decreases. Track whether people report learning or just completing.

What Gratton’s Work Reveals

Her article highlights something we should acknowledge more directly: AI isn’t just a productivity tool that happens to affect learning. It’s a fundamental redesign of the conditions under which capability develops. The Cameron and Quinn research showing 79% success rates for evolutionary cultural change versus 30% for revolutionary approaches becomes newly relevant here. AI adoption is often revolutionary in its pace—but sustainable human development remains stubbornly evolutionary.

This suggests organizations face a coordination problem. AI adoption pressure operates on quarterly timelines. Human development operates on multi-year timelines. Without deliberate intervention, the shorter timeline wins every time.

Gratton asks whether we’ll “cede agency to machines” or “design for human authorship.” We’d frame it slightly differently: Will organizations invest in the implementation infrastructure required to preserve human development, or will they let efficiency metrics make the decision for them by default?

The answer will determine whether AI becomes a tool that amplifies human capability or one that gradually atrophies it—not through malice, but through the accumulated weight of countless small decisions that prioritize speed over development.

The questions Gratton raises deserve answers. Those answers require protocols, timelines, and measurement systems. The organizations that build them will develop people. The organizations that don’t will wonder why their talent pipelines went dry.