Attachment and tool loss
The Attachment Question
If the team would grieve the agent in six months, you already have an oversight problem. Ask before you roll it out.
Dr. Sarah Dyson·September 10, 2026·3 min read·469 words
Share this note

We keep discovering attachment after the license is paid.
A team names the agent. They talk to it in the stand-up. Throughput rises. Then an upgrade, a retirement or a quiet degradation arrives, and someone says (not as a joke) I feel like I'm losing a teammate. Leaders treat that sentence as color. It's data.
The risk isn't that the system has feelings. It doesn't. The risk is that people stop practicing the work of distrust.
Relatedness is a control surface
Attachment research is about what humans do with a stable other: someone (or something) that answers, that's available, that lowers the cost of being uncertain. A well-tuned agent, used daily, occupies some of that slot. Relatedness in SCARF is why a fluent helper gets a pass that a junior colleague wouldn't.
Blind trust is the operational name for that pass. Review becomes skim. Exceptions become “the assistant usually gets this.” Audit discipline collapses at the exact moment errors can scale at machine speed. You won't see that collapse in a hallucination rate. You will see it in the meeting where nobody wants to be the person who slows the new teammate down.
When the Digital Teammate Goes Quiet is what this looks like after the cutover. How to Retire an Agent Without Retiring the Team is the holding plan once procurement has already set the date. This note is earlier. It's the question you can still ask while the system is a proposal.
The question
If we turn this off in six months, will the team grieve its loss?
A shrug is information. So is a flinch.
High attachment means competence, mood and relatedness have started to organize around a system you can revoke. That's a trust concentration. Concentrations need named owners, not congratulations.
If the answer is yes, you don't cancel the rollout by default. You staff what attachment will spend:
- A monthly reality-check: one hour where the agent is off and the team still has to produce.
- A non-anthropomorphic design rule: no name that invites a relationship the contract can't keep.
- A named auditor who is allowed to look slow next to the fluent output.
Without those, the grief at decommission is a receipt.
This week
Take one agent on the docket, not the one already loved, the next one. Put the Attachment Question on the go-live checklist. Write the answer in a sentence, with a name. If the sentence is grief, schedule the ritual before you schedule the party.
The person remains the one who can be asked why. An agent that's too costly to doubt is already governing the room.
The five questions in the Ethical AI Leadership Decision Toolkit are built for this kind of page. Field notes continue in EI Leadership Insights.
Share this note
Related in this journal
- When the Digital Teammate Goes Quiet
Teams form attachments to the systems they work beside. Retiring an agent without a grief practice is how competence leaves with the license.
- How to Retire an Agent Without Retiring the Team
Decommissioning is change management, not IT hygiene. The skills, attachments and shadow processes you skip will walk out with the license.
- The Status Threat Nobody Named
When a model joins the room, status moves. Most teams treat that as attitude. It's a threat response, and it's governable.
EI Leadership Insights
Biweekly notes by email. One practice.