Who Owns the Capability You Built at Work

For as long as people have had careers, the capability they built at work went home with them. An employer owned the output and the client relationships. Nobody owned the judgement, because judgement had no storage medium outside the person who held it.

Working through a machine changes that. Every brief, correction, rejected draft and second attempt is written down somewhere, and read together they describe how a particular person approaches a problem. That description is the closest thing anyone has produced to a transferable record of professional judgement. It sits in systems the person does not control, and almost no employment contract says anything about it.

Three questions, none of them settled#

What is kept. Not the output, which everybody knows about, but the working: the prompts, the files put in, the versions thrown away.

Who can read it. The user, the administrator, the vendor, a litigant, a future employer of the administrator.

What travels when the person leaves. The output stays with the firm. Does the record of how they got there, and if so, in what form, and to whose benefit.

What the operators say#

The vendors document this clearly, and their documentation is the best available answer to the second question. It describes capability rather than practice: what an administrator is able to reach, not what any employer does.

OpenAI's help documentation for managed accounts states that administrators may access content a user submits, including prompts, uploaded files and outputs, along with conversation history, usage metadata and security settings, and can export, audit, retain and delete it. Its enterprise privacy page adds that workspace administrators can reach an audit log of conversations through a compliance API, and that administrators control retention.

Microsoft's Purview documentation describes where Copilot prompts and responses go: a hidden folder in the mailbox of the user who ran the tool, not designed for direct access by that user, holding data a compliance administrator can search with eDiscovery tools, and searchable until permanently deleted. What the person sees in the interface is not a guide to what is retained.

None of that is a criticism of either company. Both have written it down, which is more than most employers have done. The gap is on the employer's side of the line, where the question of what the record is for has usually not been asked.

The firm has a real interest, and surveillance is not it#

Organisational experience depreciates. Bongers, estimating learning curves from procurement cost across three fighter programmes, found a persistence parameter not significantly different from zero for two of the three, meaning accumulated production experience could depreciate almost completely within a year. Benkard, working on a wide-body airliner, found the data inconsistent with simple learning and supportive of organisational forgetting, with partial transfer of experience across product generations. Both measure unit cost in aircraft assembly rather than professional judgement, so they travel to knowledge work by analogy. The direction is the useful part: a firm that does not capture how its best work is done loses it, and has always lost it.

So an employer wanting to keep the record is behaving rationally. The question is what it believes it has captured.

The record is not the capability#

A log of interactions holds the decisions that were typed. It does not hold the reasons, the discarded options that were never typed because they were obviously wrong, or the sense that an answer is off before anybody can say why. That residue is the part Polanyi named and the part tacit knowledge describes. It does most of the work.

Collins, Brown and Newman argued that apprenticeship works by making expert thinking visible, and that formal schooling fails at cognitive tasks because the reasoning stays inside the expert's head. A good interaction record is an unusually direct attempt at exactly that: reasoning, externalised, in sequence. It is also a framework argued from observation rather than a measured result, and the externalised version is thinner than the thing itself.

Both halves hold at once. The record is more than any firm has ever had. It is considerably less than the person.

What to decide, on either side of it#

For whoever runs the firm:

• Say what the record is for. Capability transfer, compliance, performance management and legal discovery are four different purposes with four different retention answers. A record kept for all of them is kept for none of them.
• Tell people, in advance and in writing. The vendors say a notice is given at set-up. An employer's own position is a separate document and usually does not exist.
• Separate the log from the practice. Capturing interactions is storage. Turning them into something a colleague can use is work somebody has to be given time for.

For whoever is building a career:

• Assume it is kept. It is the reasonable default on the documentation above, and it costs nothing to work as though it were true.
• Keep your own account. Not copies of company material, which belongs to the employer, but your own notes on the judgements you made and why. That is the part no log reconstructs and the part an interview will ask about.
• Ask what travels. Before accepting that the question has an obvious answer, read what the contract says about work product and about data. Most say a great deal about the first and nothing about the second.

What would settle it#

A survey of what employment contracts and AI usage policies say about interaction records: retention, access and what happens at exit. Nothing in the graded base measures it, and the organisations holding the answer are the ones writing the contracts. Until that exists, this page is an argument about ownership supported by vendor documentation, and not a finding about practice.

Where this sits in my own argument#

If human capability is the thing that compounds, as the SuperSkills thesis argues, then the record of how capability is exercised is a new asset and nobody has decided whose it is. The firm-side version of the same question is where your AI practice lives. The cost of answering it badly is capability debt, carried by a person rather than an organisation.

Related SuperSkills research#

On what an employer can see, can my employer see what I put into AI? On retention at the level of the firm, how do you keep expertise in an organisation? On what cannot be written down, tacit knowledge.

Key sources

• OpenAI (2026). Data access for your managed ChatGPT account.
• OpenAI (2026). Enterprise privacy at OpenAI.
• Microsoft (2025). Learn about retention for Copilot and AI apps.
• Bongers, A. (2017). Learning and forgetting in the jet fighter aircraft industry.
• Benkard, C. L. (2000). Learning and Forgetting: The Dynamics of Aircraft Production.
• Collins, A., Brown, J. S. and Newman, S. E. (1989). Cognitive apprenticeship.

About this research#

Written by Rahim Hirji, author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company.

How this research works · Reviewed quarterly · Found an error? Tell me and it is corrected on the page.