In an auspicious combination of timelines, organizations have begun to automate work with AI just as experienced employees take decades of operational knowledge into retirement. The collision could sever the path by which younger workers develop the judgment and troubleshooting instincts needed for jobs that AI is already reshaping.
An APQC survey of 1,000 organizations found only 8% consistently capture knowledge from departing retirees, while 16% make no attempt. Meanwhile, more than 11,000 Americans will turn 65 each day through 2027.
The problem is particularly acute in technology and industrial operations, where AI is assuming routine work while changing the roles successors are preparing to fill. Employers must preserve the reasoning behind critical decisions without carrying obsolete workflows into a workplace built around different tools.
Retirement Takes the Judgment Manuals Miss
Michael Morris, global head of platform and talent at Randstad Digital, says the most vulnerable knowledge tends to surface when a familiar process fails or an employee encounters an exception the documentation never anticipated.
“The knowledge most at risk usually isn’t what’s written in the manual,” Morris says. “It’s the judgment that experienced people have built over years.”
That judgment includes knowing which shortcuts are safe, who must be consulted on a difficult decision and why a system was designed a certain way. An experienced architect may eventually write less code, yet become more valuable for understanding how systems interact, where they fail and which trade-offs matter when AI handles more execution.
Jay Allardyce, CPO at industrial tech specialist Octave, says he sees the same risk in heavy industry. Manuals and blueprints can preserve standard procedures, but they rarely explain how machinery behaves in extreme weather, which in-spec readings indicate a developing problem or how an operator should approach an unusual edge case.
“It’s the very important ‘why’ behind the decision making process that a technical manual won’t supply,” he says.
Employers need to map where expertise is concentrated before a retirement date creates an emergency. Security decisions, system architecture and the rationale behind operational controls deserve particular attention because replacing a task does not replace the knowledge required to supervise it.
Capture Expertise While the Work Is Happening
Morris says a late request for a veteran employee to write down everything they know is unlikely to reveal expertise that has become instinctive.
“The mistake companies make is asking someone who has been doing a job for 20 years to sit down and ‘document everything they know,’” he says. “That rarely works because much of their expertise has become instinctive.”
A stronger process follows experts through difficult cases: diagnosing unfamiliar problems, reviewing mistakes and responding when a standard procedure breaks down. Questions should focus on specific incidents, warning signs and consequences.
Allardyce says knowledge capture must fit the employee’s normal workflow. Asking a construction worker to stop and document decisions on a tablet introduces friction; software can instead capture context while employees annotate digital twins, troubleshoot equipment or review operating data.
“Enabling AI to capture that situational context with no extra steps needed from the operator will remove administrative friction and allow teams to query operational histories while they’re working,” he says.
Kevin Spektor, co-founder and CTO at Coddy, recommends recording conversations between veteran and less-experienced employees, including candid reviews of projects that failed and how the team recovered. Mistakes often expose dependencies absent from the official process.
“Ask questions like: ‘What do you wish you’d have done before?’ or ‘What mistakes seem to be unavoidable for newer team members?’” Spektor says.
AI can transcribe those discussions, identify recurring themes and turn interviews into searchable training material.
AI Can Preserve Knowledge Without Freezing the Past
AI cannot determine on its own whether captured expertise remains accurate, relevant or safe —the APQC warned that weak content management can produce redundant or inaccurate answers, making ownership, validation and lifecycle management essential to an AI-supported knowledge program.
Morris suggests employers treat AI as a distribution mechanism for expertise rather than a substitute for the people who developed it.
“I would be careful about thinking that AI replaces the expert,” Morris says. “The real opportunity is to use AI to capture and distribute human expertise more effectively.”
Each knowledge asset needs an owner and a review date. Companies should separate durable principles, such as security constraints and architectural dependencies, from repetitive steps that automation may eliminate.
Spektor says long tenure alone does not make a practice worth preserving. Employers should ask what problem a process solves, which risks it controls and whether the same approach still makes sense after AI changes the work.
“You want to maintain the judgement and lessons that are still crucial, without turning yesterday’s workflow into tomorrow’s rulebook,” Spektor says.
Knowledge Transfer Must Be Tested on the Job
Course completion and documentation counts do not prove that another employee can handle the role. Morris recommends using simulations to test whether a successor can diagnose problems and respond when conditions change.
“I wouldn’t measure knowledge transfer by how many documents someone produced,” Morris says. “I would measure whether the next person can actually do the job.”
Allardyce points to speed to proficiency and real-world execution as stronger indicators, including how long new engineers need before they can execute an operational response without direct supervision.
“One of the clearest metrics here is low-confidence escalation success,” Allardyce says.
Other useful measures include escalation rates, repeated errors, rework and whether employees can explain why they chose a particular response.
Automation Cannot Eliminate the Path to Experience
Knowledge transfer works best as an exchange: Veteran employees contribute pattern recognition and lessons from failure, while younger workers may bring greater familiarity with AI tools.
Morris says he expects apprenticeship and rotational models to regain importance because they expose employees to real decisions. The need becomes more urgent when companies automate the junior assignments through which employees once learned how systems behave.
Employers that remove those tasks need deliberate replacements, including supervised projects, simulations, rotations and progressively harder decisions that build judgment before the organization loses its experts.
“That is particularly important as AI absorbs more entry-level tasks,” Morris says. “If the traditional junior work disappears, employers need to intentionally create new ways for people to acquire the experience that used to come from doing that work.”
