AI and the Apprenticeship Problem
Photo by Lucas von Oort - Unsplash
Designing for productivity that builds expertise instead of bypassing it
There is a growing concern that AI will quietly erode on-the-job learning for junior professionals. The concern is valid, but not inevitable. We're still early enough in the design and deployment of enterprise AI to decide whether these tools just accelerate output or build the skills and expertise that develop future leaders.
AI’s access to oceans of knowledge is extremely valuable. A marketing associate can draft a decent blog without really knowing their customers. A first-year analyst can generate a working financial model without researching the sector. A junior associate can compile a legal brief without exploring how doctrine bends.
The gains are real. But so are the gaps.
The marketing associate sends the blog to the marketing director, who immediately knows that their competitor’s solution has a clear advantage in this scenario. She knows because when she was writing content, she had to interview customers to understand why they chose this software and what tradeoffs were considered.
The first-year analyst’s model looks airtight until the finance manager asks about primary suppliers. Turns out, a dominant provider is based in a country that just entered conflict. For the manager, after years of watching supply chains fluctuate (and completely fracture during COVID), supply is one of the first things to examine when building a model.
The junior associate submits three strong legal precedents. The partner notices one case was decided at summary judgment, not their motion-to-dismiss stage. The facts look nearly identical, but the standard is different. He learned early on that arguments that win later can collapse at the outset, so he always checks posture before precedent.
The problem isn’t that the director, manager or partner caught the mistakes. That’s the win. The problem is that they only knew to look because they once did the legwork themselves.
“…they only knew to look because they once did the legwork…”
Apprenticeship, the gradual accumulation of judgment through repetition, correction, and responsibility, is how entry-level employees become decision-makers. If AI leapfrogs that formation, today's junior employees could lack the instinct required of tomorrow's leaders.
At scale, that becomes more than an individual gap. Directors, VPs, managers, and chiefs who built their judgment wrestling with incomplete information, flawed drafts and poor assumptions are more aware of pitfalls and landmines. Without that friction, institutions may find their succession pipelines short on navigational instinct.
This isn’t inevitable. Nor do we have to choose between formation and productivity.
With so much information to reference, AI's acceleration doesn't remove the need for human judgment, it increases it. The ability to clarify intent, vet assumptions, and consider variables is critical.
The way AI threatens apprenticeship is if we let it jump straight to the end. Consider a different approach.
Marketing
Instead of the associate prompting the AI to "write a blog," she opens a new project and is asked to establish intent first. What is the goal of this piece? Who is it for? What pain point are we addressing? What differentiates us here? Do we have a customer we can reference?
Her answers shape the draft the system produces. Then, after the draft is generated, the system turns the questions back outward: Does this align with the stated goal? Does it speak in the voice of the intended reader? Is the call-to-action consistent with the desired outcome?
She's not just filling in a template. She's learning what a good brief looks like before she sees one.
Finance
Before the model builds, the system asks the analyst to surface his assumptions. What are the primary revenue drivers? Where is supply concentrated, and what exposure does that create? What external risks could materially alter this projection?
The system identifies regulatory and geopolitical exposures tied to key suppliers and requires the analyst to assess which could affect the model and how. After the spreadsheet populates, the AI asks: Which assumptions drive the outcome most? Which are most fragile? What would make this model materially wrong? How confident are you in these results, and why?
The spreadsheet doesn't teach about supply chains. Exploring them before the model can proceed, does.
Legal
Before generating case summaries, the system establishes procedural clarity. What posture are we in? What standard governs at this stage? Who bears the burden? What elements must be established? Which facts are likely to be contested?
After the draft: What are the weakest points of this argument? The strongest? Are we answering the precise legal question before the court?
The procedural difference isn’t flagged by AI. The junior associate finds it when the system prompts him.
In each question answered, the junior-level employee is vetting nuance, weighing strategy, and learning what matters when and where. Having to steer builds the ability to drive. Having to rule builds the ability to judge.
Execution is still dramatically accelerated. Research is broader, drafts are cleaner, and work is produced faster. But the acceleration occurs after critical considerations have been evaluated. And as a result, the output is more focused, purpose-driven, and effective.
Employee development has never been about logging hours. It’s about learning, sometimes painfully, which questions to ask before trusting an answer.
