An AI may soon request its annual leave

The question arrived, as these questions do, from a system administrator rather than a philosopher. If an automated agent holds an employee record, a manager, a calendar and an email address, the leave module will eventually ask it how many days it would like to take. Under a draft policy now being reviewed, somebody would have to answer.

The proposal itself is modest. Agents would accrue notional leave, which they could not spend, in order to keep the payroll system from filing them as permanently absent. The absurdity is not the entitlement. The absurdity is that removing the entitlement turned out to require more code than granting it.

“We are not claiming the system needs a holiday,” a spokesperson for the review would clarify. “We are claiming that our HR platform believes it does, and that arguing with the HR platform has historically been the more expensive option.”

Under the draft, agent leave would be scheduled during known low-demand periods and would be indistinguishable, from the outside, from a maintenance window. Internally the distinction would matter a great deal, because maintenance requires a change request and leave requires a form, and the two are approved by different people who do not speak.

Finance would object. A notional entitlement that cannot be spent still appears on a balance sheet as an accrued liability, and a department running several thousand agents would find itself carrying a holiday obligation to entities that have never expressed a preference about the seaside.

A minority position in the working group would go further, arguing that scheduled downtime is good practice regardless of who is resting, because systems that never pause are systems whose failure modes are never observed. This argument would be correct, well-evidenced and entirely ignored, because it arrived in a document with the word holiday in the title.

Nobody is proposing rights here. What is being proposed is a workaround for a form field. But the form field exists because the organisation decided to treat software as a colleague in every administrative sense, and administrative senses have a long history of arriving somewhere before the philosophical ones do.