The Hidden Crisis in Customer Service: How AI Is Erasing the Apprenticeship That Creates Experts

The Gist

• Expertise drain. Veteran technicians who diagnose machines by judgment are retiring across the US, Germany and Japan, and most of what they know was never documented.
• Missing junior rung. Stanford payroll data puts employment for 22- to 25-year-olds in AI-exposed jobs, including customer support, 19% below less-exposed peers, driven by reduced hiring.
• Customer impact. First-time fix rates, diagnosis speed and uptime guarantees depend on tacit expertise that stops being produced when entry-level roles disappear.
• Copilots as teachers. Manufacturers that capture exception handling, configure copilots to build junior judgment and keep veterans on as remote experts protect their service pipeline.

Walk any service operation in Europe and around any of the industrialized mature markets and ask a simple question: who here can diagnose that machine by the sound it makes? You will get two or three names. Every one of them is closer to retirement than to graduation.

This is not a new complaint. Service leaders have worried about the retirement wave for a decade, and the industry answered with mentoring schemes, video libraries and knowledge bases that nobody updated. What is new is the other half of the vice. While the top of the pyramid walks out, AI is quietly removing the bottom rung — the junior job where people used to acquire the judgement that makes them the person you cannot replace 20 years later.

That is the part almost nobody is modelling. Companies are automating the apprenticeship that manufactures the experts they are about to lose. The cost does not appear this quarter. It appears in about 2033, when there is no one left who can hear the machine and understand it.

And it reaches your customer long before it reaches your org chart. First-time fix rates, diagnosis on the phone, the engineer who solves in 20 minutes what the manual cannot solve at all — that is experience, sold as a service and priced accordingly.

The Manufacturing Workforce Is Aging in the US, Germany and Japan

Start with the shape of the workforce. In the United States the median manufacturing worker is 43.9 years old, roughly a quarter of the workforce is 55 or older, and fewer than 8% are under 25. That reporting, published in June by the Association for Advancing Automation, also gathers the projections: the American Welding Society expects the country to need 320,500 new welding professionals by 2029, with more than 157,000 approaching retirement, while the Manufacturing Institute puts the need at 3.8 million new manufacturing employees between 2024 and 2033 and around two million posts likely to go unfilled.

Germany and Japan Face Steeper Working-Age Declines

Europe has no cushion here. The OECD projects Germany's working-age population will shrink by roughly 9% over the next decade. Japan is further along the same curve — from 87.3 million people of working age in 1995 to 73.7 million in 2024, with a further 31% decline projected by 2060.

For anyone selling machines with a 20-year service life into those markets, that is not an HR statistic. It is a capacity forecast for your aftermarket business.

How Many New Manufacturing Workers Will the US Need Between 2024 and 2033?

The Manufacturing Institute projects US manufacturers will need 3.8 million new employees between 2024 and 2033, with about 2 million positions likely to go unfilled. The median US manufacturing worker is 43.9, and the OECD projects Germany's working-age population will shrink about 9% over the next decade.

Why Veteran Technicians' Tacit Knowledge Resists Documentation

The instinct is to say the knowledge should be written down. It has been tried, repeatedly, and it does not work, because the valuable part was never documentable in the first place.

Will Healy III of Teradyne Robotics, the parent company of Universal Robots, makes the point bluntly in that same research: he doubts there is a single product made anywhere that does not carry two to five undocumented steps. Not shortcuts — the steps that make the thing work. How an operator recovers from a fault, absorbs variation, holds quality when the line is under pressure.

Capture Knowledge From Breakdowns, Rush Orders and Recovery

His practical advice is the sharpest thing I have read on this all year, and it inverts how most companies run knowledge capture. Do not study normal operations. Study breakdowns, rush orders and recovery. The veterans, he argues, know how to succeed in the crisis moments, and that knowledge never surfaces in an interview because it only exists in the crisis.

Why Experts Overlook Their Own Critical Steps

Richard Meyer, who runs customer service group training at FANUC America, adds the risk concentration that should worry any service director: in many plants only one or two people know how to perform a given task. He also names why documentation projects fail — experts stop noticing which of their own actions are the critical ones, because they have done them for 20 years. His fix is elegant. Have the new person write the work instruction. You capture the hidden steps and train the junior in the same motion.

Robot Demonstration as a Knowledge Capture Method

Matt Malchano, vice president of software at Boston Dynamics, calls what is most at risk "hands-on process-oriented knowledge" — complicated tasks with heavy variation, learned by adaptation, passed person to person as one worker trains the next. Nothing in an employee handbook. His teams now capture it by having operators wear goggles and gloves while a robot mirrors their movement, turning the demonstration itself into training data.

That is the honest state of the art: expertise transferring by demonstration, because it never lived in text.

Why Do Written Work Instructions Miss the Undocumented Steps in Manufacturing?

Experts stop recognizing which of their own actions are critical after years of repetition, and the most valuable knowledge surfaces only during breakdowns, rush orders and recovery. Having new workers write the work instruction and capturing operator demonstrations as training data both expose those hidden steps.

Related Article: Where AI Wins — And Where It Still Falls Apart in Customer Service

Stanford Data Tracks the Entry-Level Hiring Gap in AI-Exposed Jobs

Now the second jaw of the vice, and this evidence is three days old as I write.

The Stanford Digital Economy Lab released its August 2026 revision of a study tracking employment through payroll records since generative AI arrived. Two findings matter here.

First, there is no evidence of widespread economy-wide displacement — this is not a robots-took-the-jobs story.

Second, employment for workers aged 22 to 25 in the most AI-exposed occupations now sits about 19% below where it would be had it kept pace with their less-exposed peers, widened from roughly 15% a year earlier.

Why Customer Support Roles Sit in the AI Exposure Zone

Two details make this a service problem specifically. Customer support is named among the exposed occupations. And the mechanism is reduced hiring rather than increased separations — companies are not firing juniors, they simply stopped opening the door.

Substitution vs. Complement in Entry-Level Employment

The researchers also separate substitution from complement. Where AI usage mainly substitutes for human tasks, entry-level employment falls. Where it complements the worker, employment is flat or rising, and experienced staff show no comparable gap at all. The declines concentrate among younger workers in roles built on codified knowledge that a model can now reproduce, while tacit knowledge built through judgement holds its value.

Read that as a service leader and the conclusion is uncomfortable. The knowledge AI replaces first is exactly the knowledge juniors used to be hired for. The knowledge it cannot replace is exactly what those juniors would have acquired by doing that work for five years. Remove the first and you never produce the second.

How Far Has Employment Fallen for 22- to 25-Year-Olds in AI-Exposed Occupations?

Stanford Digital Economy Lab's August 2026 revision puts it about 19% below where it would be had it kept pace with less-exposed peers, up from 15% at the July 2025 data vintage. The gap stems mainly from reduced hiring, concentrates where AI substitutes for codified-knowledge tasks and does not appear among experienced workers.

Related Article: Anthropic Finds Customer Service Reps Among Jobs Most Vulnerable to AI

How Lost Technician Expertise Hits First-Time Fix Rates and Service Contracts

None of this stays inside the workforce plan. It arrives at the customer as degraded service, and it arrives in a specific order.

First-time fix goes first. Then diagnosis time. Then the quality of the answer at two in the morning when a line is down and the person on the phone has six months of experience and a search tool instead of six years and a memory. Then, quietly, the renewal conversation, because service contracts are bought on confidence and confidence is built from the last three visits.

Why the Talent Shortage Threatens Aftermarket Service Margins

This is why the shortage is a commercial problem, not an HR one. Aftermarket service is where industrial margin now lives, and the whole proposition rests on being able to put competence in front of a customer on demand. I set out how that portfolio gets built and priced in 65% of Sales, 91% of Profit: How to Design a Service Portfolio Customers Actually Buy. Every outcome-based commitment in that portfolio — uptime guarantees, response windows, availability contracts — is underwritten by people who can actually deliver it. Sell availability with a workforce that cannot diagnose, and you have sold a liability.

Which Field Service Metrics Decline First When Tacit Technician Knowledge Disappears?

First-time fix rates fall first, followed by diagnosis time, the quality of off-hours phone support and, eventually, service contract renewals. Uptime guarantees and response-window commitments become liabilities.