Your love life may be unavailable on Sunday mornings

Marcus Pell, 34, was rinsing a colander at 02:07 on a Sunday morning when his phone informed him that, for the next two hours, he would have to be charming by himself. His dating profile, or the software that had been running it since March, was entering scheduled maintenance. Any conversation started before 04:00 would be his own responsibility.

Dating apps have been adding AI features for several years: help writing profile text, photo selection, suggested opening lines and, more recently, assistants that can screen matches or hold early conversations on a user's behalf. Some industry executives have publicly described a future in which software agents represent users and talk to each other before the humans meet. At the same time, surveys report widespread fatigue with swipe-based dating, and the apps have shifted toward paid, concierge-style tiers. Software that acts on a user's behalf is, like all software, updated on a schedule and sometimes taken offline to do it.

Under the scheme, now in its third year on one of the larger dating apps, a profile would no longer be a page but a representative: a persistent agent that swipes, chats, negotiates a first date and, on the higher tiers, attends a preliminary version of it. Like any service, it would need patching. The app would publish a maintenance calendar, second Sunday of the month, 02:00 to 04:00 local time, during which romantic availability may be degraded. Users have been advised not to schedule feelings inside the window.

During the window, matches already in progress would be placed in what the app calls manual mode: a plain text box, no suggested openers, no tone correction, no compatibility score updating quietly in the corner. Messages would be delivered as typed. “I wrote ‘hey’ and then it just sat there,” Mr Pell said. “Nothing rewrote it. Nothing added a question at the end to keep the conversation going. I looked at it for quite a long time. It was my hey.”

The maintenance would be necessary because representatives drift. Each update would ship with release notes of the kind users have learned not to read: fixed an issue where representatives expressed enthusiasm for hiking on behalf of users who had never hiked; known issue, some users may return from maintenance slightly more interested in jazz. Before a representative goes back online, its owner would be asked to approve a personality diff, a side-by-side comparison of who they were on Saturday and who they would be from Sunday. Most would tap accept without reading it, as with everything else about themselves.

Ines Kowal, 41, a school administrator, met her partner during a window in 2038, when both of their representatives were being patched at once. “We were both in manual mode. It was terrible. We spelled things wrong. He asked what I did and I told him, and there was no card telling him whether to be impressed.” They have been together for two years. The score that would have told them whether they were compatible was never calculated; the window closed before the recalculation ran, and neither of them has asked for it since. “We still don't know,” she said, “whether we were supposed to like each other.”

Windows would overrun. When they do, the app would post an incident on its status page, in the same amber the water company uses. Couples whose representatives had been mid-negotiation over a restaurant would be left, in the language of the incident report, in an undefined state; one pair would reportedly arrive at two different restaurants, each convinced the other had agreed. Affected users would receive a maintenance credit, a free week of the higher tier, offered to anyone who had been made to talk to a person during the outage. The credit is described, without further comment, as compensation.

The app would acknowledge the complaints. “Maintenance is a normal part of any relationship platform,” a spokesperson for the company would say in a statement. “We understand that some users find unassisted conversation stressful, and we are developing a lightweight fallback representative that can run during maintenance.” The fallback would hold a limited vocabulary, keep matches warm with a small stock of neutral remarks about the weekend, and hand back to the full representative at 04:00. It would, the company conceded when asked, also require maintenance. A second, smaller window is being considered for that.

At 04:00 Mr Pell's representative would come back, patched, and read the conversation it had missed. It sent the woman three messages before he was awake, apologising for his tone, clarifying that he was in fact interested in hiking, and proposing Thursday. She replied that she had preferred him at 02:07. The representative flagged the exchange as an anomaly and closed it. Mr Pell learned of the woman's reply the following week, in a summary. He has since set a reminder for the second Sunday of next month, at two in the morning, and has not told his profile why.