How do we learn without doing?

The apprenticeship model in knowledge work will need a fundamental reboot—and soon.

One of my early mentors in consulting, Bob Lurie, PhD. , stressed that one of the most important tasks in learning—for our own people and our clients—was showing people “what ‘good’ really looks like.” No matter how good the processes and tools are, you need to know what a great outcome or output looks like, and why it is indeed “good.” We typically learned that by grinding away at the work, taking on bigger and bigger pieces, until we had built the requisite judgment and discernment.

We’re on the precipice of a world where GenAI-based platforms and agents can instantly generate thousands of versions of “good,” nearly instantly, for most knowledge work. The entry-level “doing” roles will shrink, and those staff will instead be tasked with managing and using those platforms and agents. So how will we quickly teach judgment, discernment, taste and pattern recognition? We believe a radical re-boot of the traditional training and development approaches will be needed, and we’ll need that reboot faster than we think.

?????? ?????? ????????????????: ???????????????????????????? ?????? “???????????????? ???? ??????????”

I grew up professionally as a consultant. As a new analyst, I largely learned by “doing”—building a financial or analytical model, selecting the right charts to display its outputs, and distilling the key message into a single slide or two that fit into a broader storyline. I got feedback (quickly and voluminously) on what was good in my work, and what needed improvement. Over time, I gained exposure to more complex problems, worked directly with clients, and took on larger and larger parts of the engagement. In doing so, I developed a deeper understanding of the nuances of strategy, built pattern-recognition and learned how to create change with individuals and within organizations (or at least got better at it).

More broadly, most knowledge work industries have historically used an apprenticeship model to develop their talent. Entry-level employees are typically given low-risk, foundational tasks, which may seem mundane but were critical in both delivering the work and developing skill. In marketing, young professionals often learned by working on discrete parts of a campaign, such as selecting fonts, writing copy, or curating images. Through these experiences, they gradually developed a keen sense of what makes a campaign successful, learned to recognize good design, and honed their skills in targeting specific audiences. Similar paradigms exist in fields like law, finance, and design: early-career employees learn by doing.

?????????????? ???? ?????? ?????? ??????????-??????????

However, as GenAI continues to take over these tasks—particularly the repetitive, time- consuming, and labor-intensive ones—new hires will no longer be expected to perform them. Multiple GenAI platforms can now design entire marketing campaigns and generate sophisticated analyses and reports at a fraction of the time and cost it would take a human team. In consulting, this shift is already underway, with Lilli at McKinsey and BCG’s enterprise GPT in use across the firms in co-pilot modes.

While these capabilities offer enterprises tremendous efficiency and can elevate the average quality of work done, it raises an important question: How do we train new workers when entry-level tasks are increasingly being handled by machines?

???????????????? ???????????????? ?????? ?????????????????????? ???????? ???????? ?????????????? ???? ???????????????? ????????????????

As AI takes over execution, the new skill set for entry-level workers will involve managing, interpreting, selecting and elevating AI-created outputs. This will require training and assessment approaches that emphasize critical thinking, discernment and judgment instead of task-mastery.

In consulting, instead of creating individual slides, the new consultant might be asked to compare AI-generated presentations, identify the most persuasive narrative, and modify it for the client’s specific needs. Similarly, a marketing or design interview typically includes reviewing the applicant’s portfolio of past work. But soon, evaluation may evolve into modes like, "Here are three AI-generated campaigns. Which one best aligns with our brand’s voice, and why?" Training and assessment processes will shift from "learning by doing" to "learning by evaluating, testing and refining."

???????????????? ?????????????? ??????????

To prepare for this AI-driven future, enterprises need to consider new training paradigms:

1. ????????????????????-?????????? ????????????????: Simulations are great tools for developing contextual decision-making and behaviors, and can also compress and distill years of experiences into days. Simulations featuring scenarios where AI tools produce work and employees must evaluate, refine, and explain their decisions will be a critical training modality. Related games and logic puzzles can also help build pattern recognition more quickly.

2. ?????????? ???? ????????????????-???????????? ?????? ????????????????: Training programs will need to prioritize decision-making under uncertainty. Exercises in scenario planning, problem diagnosis and structuring, risk assessment, cognitive biases and other approaches will all become essential and need to be delivered sooner in the employee journey.

3. ???????????????????????????? + ????: Traditional apprenticeship models will still have value, but the focus will shift to teaching new hires how to interpret AI outputs, iterate with those tools to drive deeper work, and apply their own expertise to enhance AI-generated solutions. Mentors can guide more junior employees on how to spot when the AI’s conclusions are inaccurate or miss subtle contextual nuances.

4. ???????????? ?????? ???? ????????????????: New hires will also need a strong foundation in AI literacy. Understanding how AI models work, their limitations, and their ethical implications will be needed to reduce over-reliance on technology and ensure responsible use.

Until our machine overlords are ready...

Over time, as outcomes and results are fed back in closed-loop systems, GenAI may eventually have a superior sense of “what good looks like.” Until then, that’s still our job as humans and one we need to find a way to get quickly get better at. Would love to hear from any leaders who are grappling with this issue and how you’re approaching it!