In August 2025, I shared a perspective on LinkedIn about AI changing leadership pipelines and organizational design. The idea that stayed with me was the effect on how people learn to run a business.
Junior employees often develop judgment through work that looks routine from a distance. They prepare an analysis, investigate a discrepancy, draft a customer response, and learn why an experienced colleague changes it. AI can accelerate parts of that work. Leadership still has to decide how the learning will happen.
I would include that question in an AI transformation plan. A company can improve near-term productivity while making it harder to develop its next generation of managers.
Identify what the task teaches
Before redesigning a role, ask which skills the current work develops. Preparing a weekly report might teach someone where the data comes from, what makes a number unreliable, and which changes deserve investigation. Drafting a proposal might teach the relationship between scope, customer expectations, and margin.
Some manual steps teach very little. Copying the same information between systems is a good candidate for removal. Other steps expose people to decisions they will eventually need to make independently. Treating both categories as interchangeable hours misses that difference.
The manager should be able to name the judgment an employee needs to acquire and the work that will develop it after the process changes.
Ask employees to explain the recommendation
An AI-generated analysis can become a useful learning tool if the employee must test it and explain it. Ask what evidence supports the conclusion, which assumptions matter, and what would change the recommendation.
For a hypothetical analyst preparing a customer profitability review, I would have them compare the output with source records, explain how returns and support costs were treated, and investigate an unusual account. That work develops commercial understanding even if the first draft arrived quickly.
Approval alone is a weak learning mechanism. An employee can click through a review without understanding the business. The conversation with the manager is where a great deal of the development happens.
Keep exposure to customers and exceptions
When routine work is automated, newer team members may encounter fewer customers or less of the operating process. Deliberately replace that exposure.
Let them observe difficult customer conversations, join an operating review, investigate a failed handoff, or own a bounded improvement. Pair them with experienced people who can explain the reasoning behind a decision. Give them responsibility appropriate to their ability and review the result.
This takes management time. Include it when evaluating the productivity benefit. The goal is a stronger team that can perform the work and improve it.
Update roles before measuring them
A role that previously rewarded output volume may need to reward investigation, problem resolution, and process improvement. Make the expectations clear. Employees should know whether they are responsible for drafting, validating, deciding, or acting.
Ask managers whether the remaining work is more demanding. If the easy cases disappear, an employee's lower transaction count may reflect a harder workload. Training, supervision, and performance measures need to follow the actual job.
For a portfolio company, succession is part of the operating asset. Reducing repetitive work can be valuable, but the business still needs people who understand customers, economics, and tradeoffs.
Make development part of the deployment decision
I would ask each workflow owner to answer two questions: how will this change improve performance, and how will people learn the judgment needed to supervise it?
That second question does not require preserving unnecessary manual work. It requires a deliberate plan for developing capable people as the work changes. In my experience leading teams, a tool gets much more useful when the people using it understand the business well enough to challenge its output.
Source note: This new article develops a theme from my LinkedIn share dated August 13, 2025. It is published on this website in October 2026.
Related reading: Controls for AI agents and Choosing fractional technology leadership.