Learning in the Flow of Work: How AI Compresses the Path from Novice to Expert

For the First Time in History, Apprenticeship Has a Metric: the Narrowing Gap Between the Employee's Attempt and the AI's Output

Building Expertise in the Age of AI (3)

Apprenticeship has always had a mysterious quality. You knew when it worked—the junior grew into an expert—but you could rarely say precisely when the shift happened or measure it in the moment. That is one of the reasons organizations have quietly disinvested in it over the years: what cannot be measured tends to lose funding. In the AI era, that changes. As McKinsey observes in Building expertise in the age of AI (Hancock & Seiler, July 2026), the answer-key model even supplies the metric. The gap between an employee's independent attempt and the AI model's output is observable, and a gap that narrows over time is direct evidence that judgment is forming—something apprenticeship, for all its history, has never been able to measure.

Play 3 makes that measurable growth continuous. It embeds learning directly into daily work through what the article calls learning in the flow of work. Employees are not waiting to become productive. They are producing real outputs while AI acts as embedded guide and reviewer—and their judgment is being measured, in real time, as they go.

From Training Program to Learning System

Traditional corporate learning was episodic. Employees stepped away from the work, sat in a program, learned something in the abstract, and returned. It rarely stuck. Learning in the flow of work reverses the relationship. The work itself becomes the learning environment. AI tools suggest next steps, flag risks, and provide instant feedback as the employee produces. Managers stop teaching task mechanics (the AI handles that) and focus their coaching on judgment and decision-making. The training program does not go away. It becomes more purposeful—front-loading concepts that need context before doing—while the deeper skill-building happens in the actual doing.

Two Design Patterns

The article observes two dominant patterns emerging:

• Front-loaded onboarding — short simulations and structured programs that establish a shared mental model before the employee touches real work

• Fast entry into real work — trusting that AI-enabled feedback loops and coaching will accelerate development in the moment

Either can succeed. In both, the underlying model is the same: employees are not waiting to become productive; they are learning by doing, with AI compressing the time it takes to build experience and earn trust.

Why the Attempt-Then-Check Loop Is Now Backed by Learning Science

The design principle underneath—attempt first, then check against the AI—is beginning to accumulate serious evidence. In one experiment, first-year medical students who answered AI-generated case questions and received automated, personalized feedback over five days outperformed second-year students, a full year ahead in training, on the targeted diagnoses in a video-based exam—including at a two-week follow-up. The learning persisted. That last detail matters. It provides evidence that the attempt-then-check loop can produce durable skill, not borrowed competence.

A Stanford pilot pushes the design further: a chatbot plays the patient, and the student interviews the patient, commits to a diagnosis, and defends it—only then does the system critique the reasoning. The sequencing is the whole point. The student goes first. AI grades the attempt. The comparison is where judgment forms.

The Gap Is the Signal

Here is the quiet revolution of learning in the flow of work: for the first time, organizations have a real-time signal of skill development. The gap between a junior employee's independent output and the AI's output can be measured on every meaningful task. As that gap narrows over months, judgment is forming. When it widens on a new type of problem, the employee has hit a growth edge worth coaching. When it disappears entirely, the employee is ready for harder problems. Learning ceases to be a mystery and becomes a metric. That is a change of remarkable significance for how organizations design careers, invest in people, and manage performance.

Design and Dose Carefully

One caution the article rightly emphasizes: workflows that demand this kind of cognitive engagement are harder to use and can overwhelm beginners. They must be deliberately designed and dosed—not bolted on. Some tasks should let AI move fast. Others should slow the workflow enough to force the attempt-and-compare step. Getting the mix right is a design decision, not a technology decision. Too little friction and the employee never develops judgment. Too much and the system feels punishing and adoption collapses.

What Continuous Support Looks Like Around the Employee

In a well-designed learning-in-the-flow-of-work system, continuous support surrounds the employee:

• AI systems flag risks, suggest next steps, and provide instant micro-feedback as the work happens

• Managers coach on judgment and decision-making, not task mechanics

• The knowledge base (Play 1) makes institutional expertise available in the moment

• The attempt-then-check loop is preserved as a habit, not a one-off exercise

• The narrowing gap becomes part of how performance and growth are actually reviewed

What This Means for Data & AI Leaders

Concretely:

• Build the attempt-then-check loop into the daily workflow, not just onboarding

• Treat the gap between employee output and AI output as a first-class development metric

• Design friction on purpose—slow the workflow at the moments where judgment is being formed

• Rebalance manager time from supervising task mechanics to coaching judgment and decision-making

• Front-load learning where context is needed; move to fast real-work entry where AI feedback can carry the load

A Moment of Reflection

Sit with these:

• Where in our daily workflow does the employee attempt first—and where do they immediately reach for the AI?

• Are we measuring the growth of judgment—or only the output of the work?

• Have we designed enough productive friction into the flow, or has AI made everything feel too easy to build any real skill?

For the first time, apprenticeship has a metric and a rhythm that scale. Learning in the flow of work turns every real task into a chance to build judgment, and every narrowing gap into visible proof that the next generation is becoming what the organization needs it to become. The winners of the AI era will not simply have better models. They will have designed better learning loops around them. 🌊

In Play 4, we turn to the front door: hiring for judgment, not tool fluency—and why the durable bet on entry-level talent is now the "general athlete" over the narrow specialist.

Stay tuned for the next blog, and subscribe to the blog and our newsletter to receive the latest insights directly in your inbox. Together, let's make 2026 a year of innovation and success for your organization.

>> Discover the path to achieve sustainable growth with AI and navigate the challenges with confidence through our Data Science & AI Leadership Winning Blueprint that's tailored to help you craft a compelling data and AI vision and optimize your strategy—it's your key to success in the journey of Generative AI. Reach out for a complimentary orientation on the program and embark on a transformative path to excellence.

May you grow to your fullest in your data science & AI!

Subscribe Grow to Your Fullest and

Get Your FREE data & AI Leadership Blueprint, or

Book a FREE strategy call with us

Learn more Data & AI strategy consulting framework