Every team has one: the person who remembers what was discussed three meetings ago, notices when someone goes quiet, and makes sure the right people talk before things fall apart. This work has no job title. It doesn't show up in sprint velocity or quarterly metrics. In organizational research, it's called "glue work" - the connective labor of relationship-building, context-sharing, and coordination that binds teams together.
A recent essay by software tester Patrick Prill makes a sharp argument that this is precisely the kind of work AI tools should not be automating, even when they technically can.
The Contact Surface Problem
Prill borrows a woodworking metaphor to make his point. Joint strength depends on surface area - the more contact between two pieces of wood, the stronger the bond. Organizations work the same way. When someone walks to another desk to ask a question, that interaction isn't just information transfer. It's relationship maintenance. It builds the kind of ambient trust that makes teams function under pressure.
Automate that interaction with an AI summary bot or a Slack integration, and you get the information faster. But you lose the contact surface. Do it enough times across enough interactions, and the organizational joints start weakening - "just with a quiet crack when you put weight on it," as Prill puts it.
This isn't a Luddite argument. It's a structural one. The act of gathering information was itself doing work that no dashboard can replicate.
The Invisibility Trap
Here's where it gets uncomfortable for anyone managing AI adoption: glue workers are the most likely to be automated away because their contributions are the hardest to measure. They often show lower individual output precisely because they're spending time on coordination that benefits everyone else. When an AI tool promises to "handle" meeting summaries, status updates, and cross-team communication, the people who lose their roles first are the ones who were doing that work informally.
The degradation isn't immediate. Teams don't collapse overnight when you remove a glue worker or automate their tasks. They become slower, more brittle, more transactional. Misunderstandings that used to get caught early now escalate. Context that used to flow naturally now lives in tools nobody checks.
Prill describes it as losing the ability to "read the room" - detecting when trust is eroding, when someone is struggling, when two teams are about to collide. AI can summarize a meeting transcript. It cannot notice that someone's tone shifted in a way that means trouble.
A Practical Warning for Tool Adoption
This matters for anyone rolling out AI productivity tools across a team. The ROI calculation on automating coordination work is usually straightforward: hours saved, messages reduced, fewer meetings. What doesn't appear in that calculation is the relationship infrastructure those activities were quietly maintaining.
The smarter approach is treating AI as a tool that frees glue workers to do more glue work, not less. Use AI to handle the mechanical parts - formatting notes, pulling data, drafting agendas - so the humans can focus on the parts that actually require being human: reading context, building trust, and keeping the joints from cracking.