---
title: When the Employee Quits, Her AI Skills Stay Behind
subtitle: AI skills can retain operational know-how after an employee leaves. That
  makes ownership, confidentiality, and offboarding concrete management questions.
description: How AI skills turn individual workflows into company assets—and why employees
  and employers need clear rules for ownership, reuse, and offboarding.
author: Mikkel Frimer-Rasmussen
last_updated: '2026-09-22'
type: TechArticle
status: published
tech_stack:
- generative AI
- AI agents
- knowledge management
- governance
primary_goal: Explain how organisations can retain operational AI skills without creating
  unclear ownership or confidentiality risks.
canonical_url: https://frimer-rasmussen.dk/articles/when-the-employee-quits-her-ai-skills-stay-behind.html
tags:
- AI skills
- knowledge management
- intellectual property
- trade secrets
- governance
concepts:
- name: AI skill
  description: A structured instruction file that lets an AI agent perform a defined
    part of a workflow using approved sources, controls, and escalation points.
- name: Skill grief
  description: The loss a professional may experience when company-owned AI workflows
    preserve part of their practical execution method after they leave.
- name: AI-skill governance
  description: The rules, repositories, ownership records, approval controls, and
    offboarding processes that govern reusable AI workflow files.
---

# **When the Employee Quits, Her AI Skills Stay Behind**

**When I left my last job, my skills left with me inside my head. If I did that today, a significant part of them would remain behind as markdown files.**

 

I became an independent consultant to work more freely with AI. Back then, my workflow existed mostly as experience, daily habits, and professional judgment. It was not saved in files that other people could execute.

 

Today, I have more than 40 skills. These are plain text files that instruct an AI agent how to solve a given task: which sources to consult, what steps to take, which errors to look for, and when a human needs to evaluate the case.

 

For instance, a skill can direct an agent to review tender documents, extract binding requirements, flag uncertainties, verify output against specific sources, and deliver the result in the company's preferred format. The file merges the employee’s method with company quality criteria, internal systems, concrete examples, and known pitfalls.

 

A standard operating procedure tells a human worker how to do the job. An AI skill directs an agent to actually run a defined part of it. Tasks that once required a specific specialist can now be handled by AI inside company systems and then reviewed by a team member.

 

The skill does not capture the specialist’s complete domain expertise. A person still needs to select the approach, spot edge cases, and take responsibility. Yet the file captures and retains far more practical day-to-day execution than a traditional job description or handover document ever could.

 

## **Employees Started Without Leadership**

On September 16, 2026, the Danish think tank Mandag Morgen published a study on workplace experiences with generative AI. Among respondents at Danish workplaces where generative AI is permitted or required, only 17 percent report a strategic, leadership-driven implementation. Thirty-nine percent report getting zero direction from management.\[^1\]

 

While not representative of Denmark's entire labor market, the 3,174 respondents illustrate a clear trend: staff members are restructuring work routines before management has decided how to guide the transition.

 

Mandag Morgen did not look specifically at AI skills. However, the logical next step is evident: when employees adopt AI tools autonomously, some begin codifying their workflows into files that AI agents can reuse.

 

It often starts with a prompt for research or proposal writing. Step by step, the employee adds company standards, verification checklists, and hard-earned lessons from past mistakes. A few months in, a substantial part of her professional workflow resides in a file—even if no manager ever asked for it.

 

## **Skill Grief**

On her last day, the employee returns her laptop and says goodbye to colleagues. Left behind in a folder are the skills refined across hundreds of micro-adjustments. They can still drive an agent to carry out parts of her former role.

 

Starting her new job, she launches the same tools, only to realize she cannot deliver with the same speed and confidence from day one. We may need a distinct term for this sensation: **skill grief**.

 

That loss hurts, even when the files legitimately belong to the company. The employee encoded her personal craft into them, often on her own initiative and without executive backing. The employer gained continuous value during her tenure and now keeps the workflows and operational capacity built for the business.

 

This dynamic tilts the balance firmly toward the company. A colleague can execute the skill without spending years accumulating the predecessor’s expertise. A newer AI model might follow the instructions even more effectively. If the skill automates enough of the workflow, the company might rethink staffing levels entirely.

 

The departing specialist still carries her experience, intuition, and edge-case mastery. But she has to rebuild and adapt everything from scratch inside the new employer's systems, data, and constraints.

 

An AI skill cannot take accountability, nor can it intuitively grasp novel contexts. It can just as easily perpetuate outdated errors. Employers still rely on human professionals to inspect deliverables and take real ownership.

 

## **Who Owns an AI Skill?**

The legal outcome hinges on the contents of the file, the employment agreement, and the exact circumstances of its creation under Danish law.

 

The Danish Salaried Employees Act (*Funktionærloven*) does not automatically assign every idea, workflow, or text created by an employee to the employer. The employment contract is therefore decisive. Many Danish contracts include clauses addressing intellectual creations, methods, and inventions produced during employment. An AI skill developed on the job, tailored for company tasks, and utilizing proprietary data is clearly a work product subject to such clauses. Using personal initiative or a private AI subscription does not inherently grant the worker ownership.

 

An original markdown file can qualify for copyright protection under Danish copyright law. This protection shields the specific phrasing and structure, not the underlying methodology as an abstract concept.\[^2\] In an employment context, rights transfer according to contract terms. Even without an explicit clause, Danish case law recognizes that employers acquire the usage rights strictly necessary for their ordinary operations.\[^3\] Furthermore, if a skill contains actual software code, Section 59 of the Danish Copyright Act (*Ophavsretsloven*) regarding computer programs may automatically transfer the copyright to the employer.\[^4\]

 

The file may also embed information regarding clients, pricing structures, system architectures, and internal workflows. Under the Danish Trade Secrets Act (*Lov om forretningshemmeligheder*), these count as protected trade secrets if they remain confidential, carry commercial value due to their secrecy, and are subject to reasonable protective measures.\[^5\]

 

At the same time, Danish and EU trade secret laws explicitly preserve the right of employees to deploy the general experience and skills acquired honestly during normal employment.\[^6\]

 

Drawing the line proves difficult because an AI skill usually weaves general logic together with client details, internal phrasing, and proprietary business choices. Stripping the logo off a file rarely makes it safe or legal to export.

 

## **Employees Must Clarify What They Can Take With Them**

Professionals should examine their contract, internal AI policies, and Danish workplace rules on confidentiality and IP before investing months in crafting skills. Five clear steps can help reduce their exposure:

 

1. **Document existing assets.** Detail any pre-existing skills, workflows, or templates brought into the role via an agreed-upon contract addendum, free of confidential material.  
2. **Establish boundaries in writing.** Clarify ownership, license terms, and the freedom to reuse generalized workflows stripped of company text, data, and context.  
3. **Separate generic workflows from company data.** Only do this with the employer’s explicit consent.  
4. **Showcase capability without copying files.** Agree on how results, methodologies, and technical achievements can be presented on a CV or in a portfolio.  
5. **Clarify departure procedures.** Secure written guidance on what files must be handed over, deleted, or kept. Never hoard file copies as personal insurance.

Operating on personal hardware or paying for a private AI subscription does not automatically make the resulting skill private property. Nor does a company's silence equate to permission to take files. Securing the right to reuse a general method requires agreement before it becomes entangled with proprietary company assets.

 

Student assistants, temporary agency workers, project staff, and freelance contractors face an even higher urgency to clarify terms upfront due to the temporary nature of their engagements.

 

A student assistant's early AI skills may represent the very first codification of her professional skillset. Fixed-term project workers know their end date and should negotiate boundaries on day one. Temporary agency workers operate under the direction of the client company while legally employed by the agency under Danish legislation governing temporary agency work.\[^7\] This arrangement demands absolute clarity on applicable policies, file storage, and ownership rights.

 

For freelancers, existing libraries, prompt chains, and skills represent core capital. Service contracts should sharply delineate pre-existing assets, client-specific deliverables, and residual rights to reusable framework logic. Delivering a work product, licensing usage, and transferring full intellectual property are three entirely separate transactions under Danish contract law.

 

## **The New Employer Can Import Legal Conflict**

A new employer often hires talent based on output driven by a blend of personal judgment, proprietary workflows, and the previous company's infrastructure. Naturally, the new recruit wants to demonstrate that same high velocity immediately.

 

Importing old files, however, can introduce copyrighted text, trade secrets, or assets legally transferred to the former employer. Even an ostensibly generic skill might conceal sensitive customer data, pricing formulas, or vulnerability logs. The hiring company risks contaminating its own systems with illicit copies, exposing both the worker and the organization to dispute.

 

During onboarding, managers should verify whether new hires are bringing AI files and ensure they have documented rights to use them. The organization hired the individual's talent and critical thinking—not a backdoor license to an ex-employer's IP.

 

## **Management Must Set the Rules**

A sweeping contractual clause is useless if team members build skills using unauthorized accounts while no one knows where the files reside. Executive leadership, IT, HR, and department heads need four coordinated decisions:

 

1. **Maintain an inventory.** Catalog active skills across the company, outlining their designated tasks, source data, and designated human approvers.  
2. **Differentiate generic from proprietary logic.** Update standard employment, temp, and contractor agreements to distinguish between baseline methods, proprietary deliverables, and permissible reuse.  
3. **Structure the offboarding process.** Centralize skills in sanctioned company repositories, track ownership and versioning, and test before an employee departs whether teammates can operate the files safely.  
4. **Screen incoming files during onboarding.** Demand documented proof of ownership. If proof is missing, require the hire to construct new skills from the ground up using internal tools, data, and criteria.

Clear frameworks safeguard corporate assets while setting clean boundaries for what employees can carry forward. They also prevent new hires from inadvertently importing legal liabilities from former employers.

 

## **Starting With a Blank File**

At her new desk, the specialist opens an empty SKILL.md.

 

She does not duplicate her old files. Instead, she relies on her experience to ask the right questions from scratch: Which tasks deliver the most value here? Which data sources are reliable? What edge cases will the agent encounter? Who validates the output? What information is the agent permitted to read?

 

Her former employer can keep running parts of the role using the skill left behind. She, on the other hand, must patiently draft a slower first iteration tailored to her new environment.

 

The new SKILL.md begins empty. She does not.

 

## **Sources**

\[^1\]: Danish think tank Mandag Morgen, *Når generativ kunstig intelligens rykker ind – medarbejdernes erfaringer med et arbejdsliv i forandring* \[When generative AI moves in: Employee experiences in a changing workplace\], September 2026, pp. 7 and 14–16. The survey covers 3,174 employees at workplaces where generative AI is permitted or required, and is not representative of the broader Danish labor market. See also [IDA's coverage of the report](https://ida.dk/om-ida/nyt-fra-ida/de-dygtigste-faar-mest-ud-af-generativ-kunstig-intelligens?utm_source=gemini).

\[^2\]: Danish Ministry of Culture, [*Ideer og koncepter*](https://kum.dk/kulturomraader/ophavsret/ideer-og-koncepter?utm_source=gemini) \[Ideas and concepts under Danish copyright law\], updated June 12, 2026\.

\[^3\]: Danish Ministry of Culture, [*Betænkning om revision af ophavsretslovens kapitel 3* (Report No. 1480/2006 on the revision of Chapter 3 of the Danish Copyright Act)](https://www.ft.dk/samling/20061/almdel/kuu/bilag/38/316625.pdf), pp. 22–25.

\[^4\]: [The Danish Copyright Act, Section 59 (Ophavsretsloven)](https://www.retsinformation.dk/eli/lta/2023/1093?utm_source=gemini).

\[^5\]: [The Danish Trade Secrets Act, Section 2 (Lov om forretningshemmeligheder)](https://www.retsinformation.dk/eli/lta/2018/309?utm_source=gemini).

\[^6\]: [Directive (EU) 2016/943 of the European Parliament and of the Council on trade secrets](https://eur-lex.europa.eu/eli/dir/2016/943/oj/dan?utm_source=gemini), Article 1(3) and Recital 14 (implemented into Danish law).

\[^7\]: [The Danish Act on the Legal Status of Temporary Agency Workers, Section 2 (Lov om vikarers retsstilling ved udsendelse af et vikarbureau)](https://www.retsinformation.dk/eli/lta/2013/595?utm_source=gemini).

 

*This article outlines general legal principles under Danish law and does not determine the specific rights to any individual AI skill. Companies should obtain independent legal reviews of their contracts and operational procedures when establishing internal policies.*