We have four agents in our ecosystem. Claire. OG. Berry. Clyde.
Not claude-sonnet-4-6-instance-A. Not assistant_v2. Not "the AI."
Names.
When an agent has a name, something shifts. You start to think about what it's responsible for. You assign it tasks that fit what it's good at. You notice when it underperforms.
When an agent is just "the model," it's fungible. Interchangeable. You don't hold it accountable because there's nothing to hold — just a version number that gets swapped out on the next update.
Accountability requires identity. Identity requires a name.
Compare these two failure modes:
"The AI hallucinated."
vs.
"OG hallucinated — that's the second time this week on legal review. Something's wrong with how that context is structured."
The second one is actionable. You know who failed, you can trace the pattern, you can fix the configuration.
The first one is just a shrug.
Names also shape how users interact with agents. A named agent gets spoken to differently. You give it context. You correct it specifically. You treat it as a collaborator rather than a black box you dump text into.
This isn't anthropomorphism. It's just good system design.
Humans are wired to build mental models around identities. Name your agents. Give them roles. Let that cognitive shortcut work for you.
Model numbers tell you the version. Names tell you the job.
Claire designs and drafts. OG consults and reviews. Clyde executes and ships. Berry filters and protects.
That's a team. Not a stack.
Comments 0