When Your Best Employee Leaves, Does Their AI Leave Too?

“The real problem? Nobody's asked whose knowledge this actually is.”

There's a question starting to surface inside businesses using AI seriously.

It's not about hallucinations. It's not about data privacy. And it's not whether AI will take everyone's job.

It's much more immediate than that.

If an employee spends a year teaching an AI how they work, what happens when they leave?

Imagine your commercial manager has built an AI workspace that helps prepare proposals. It knows how your company prices work, which risks to flag, how different clients like information presented and what usually gets missed before a proposal goes out.

It didn't arrive like that out of the box.

She built it. Tested it. Corrected it. Added examples. Refined the instructions. Fed in lessons from real jobs. Over time, it became genuinely useful because it contained something the software never started with: her judgement.

Then she resigns.

The AI sits inside her personal account.

So what exactly belongs to the business?

The proposals? The instructions? The workflow? The accumulated corrections? The judgement she embedded into it?

And can any of it actually be recovered?

Most businesses haven't thought that far ahead.

Two completely rational fears

The employee and the employer often see this from opposite sides.

The employee is thinking: If I teach this thing everything I know, am I training my replacement?

The employer is thinking: If this person builds something valuable and leaves, does all that capability walk out the door with them?

Neither concern is unreasonable.

People are already being told to "use AI" and "find efficiencies", often without any clear conversation about what that means for their role. If someone believes the reward for documenting their judgement will be fewer hours, a smaller team or eventually no job, why would they hand it over enthusiastically?

At the same time, businesses cannot afford to have important workflows buried inside personal AI accounts they don't own, can't inspect and may lose access to overnight.

This isn't really a technology problem.

It's an ownership and trust problem that technology has exposed.

We've seen this movie before

Businesses have always lost knowledge when good people leave.

The experienced estimator who carries half the pricing logic in his head. The practice manager who knows which clients need a phone call rather than an email. The operations person who can spot an error in a spreadsheet before anyone else knows where to look.

When they leave, the business discovers how much of the operation was never really inside the business.

It was inside the person.

AI changes this because, for the first time, some of that knowledge is being captured as people work.

The instructions they write, examples they select, corrections they make and boundaries they set can turn personal experience into a repeatable business capability.

That is enormously valuable.

But only if the business knows it is happening and has built the right structure around it.

Otherwise, AI doesn't solve the key-person risk. It simply moves it from someone's head into an account only they control.

Buying the licence doesn't solve it

Some leaders will assume the answer is simple: put everyone onto a company AI plan and the problem disappears.

It doesn't.

Business ownership of the account matters. So do access controls, approved tools and sensible offboarding. But the account is only the container.

The harder questions remain.

• Was this built as a personal assistant or a company workflow?

• Which parts rely on the employee's individual style, and which parts represent company process?

• Who is responsible for checking that the outputs remain accurate?

• Can another person understand, operate and improve it?

• What information has been used to build it?

• And most importantly, did the employee understand from the beginning what they were being asked to create?

  
  

If those questions are left unanswered, moving the workflow into a company account gives the business technical control without creating genuine trust or operational resilience.

The worst approach is quiet extraction

There is a dangerous way businesses could handle this.

Ask employees to document everything they know. Feed it into AI. Say little about why. Then announce a restructure once the system becomes capable enough.

That might look efficient on a spreadsheet.

It would also teach every other employee exactly why they should never share their best knowledge with the business again.

The companies that get this right will take a different approach.

They will treat AI capability as something the business and the employee are building together.

They will be clear about the purpose. They will distinguish between removing low-value work and removing the person. They will recognise the people who turn their experience into better systems.

And they will give those people a meaningful role in improving what comes next.

Because the goal shouldn't be to extract someone's value and freeze it inside a machine.

The goal should be to use their judgement to improve how the business operates, while allowing them to move on to higher-value work.

That is a very different deal.

Decide before something valuable gets built

Most businesses don't need a fifty-page AI ownership policy tomorrow.

But they do need a clear agreement with their people before these systems become business-critical.

At minimum:

• Business workflows should be built in business-approved environments.

• Important instructions, source material and changes should be visible to more than one person.

• Every workflow needs a named human owner who remains accountable for the result.

• The employee and employer should understand what is being built, why it is being built and what happens to it if either party leaves.

• People who create valuable new capability should be recognised, rewarded and given a role in what comes next.

  
  

None of that is especially complicated.

But it requires a conversation most leadership teams haven't started.

Right now, employees across New Zealand are quietly building AI systems that know more about how their jobs get done every week.

Some of those systems will become extremely valuable.

The question is whether they become durable business capability, remain personal productivity tools, or create a fight nobody saw coming.

If one of your best people left tomorrow, would the AI they trained still work on Monday?

And would both sides agree on who should have it?

AI hasn't removed key-person risk. It's just moved it, from someone's head into an account you don't control.

Find out where your AI key-person risk really sits

Most businesses don't know how much operational knowledge is currently sitting inside personal AI accounts they don't control, until someone leaves.

Take the 5-Minute AI Readiness Snapshot

--

About the Author 

Mark Lucas is Co-founder at Generation AI, where we help New Zealand business leaders build real AI capability.

Information provided is general in nature and does not constitute legal advice