After Users Gave AI Job Titles, They Started Drawing Org Charts

After Nuphos launched sub-agents, the team noticed something: users started renaming sessions to job titles. Release Engineer, Incident Manager, Code Review Engineer. No one called them “the conversation that fixes bugs,” and no one called them “that task on Wednesday afternoon.” The names were job titles, like people’s names.

They were no longer treating these sessions as chat logs. They were hiring.

So Yuanlin Lin gave each agent a face. A day later, they blinked.

In the interface, three roles sit side by side: Checkout Incident Manager wearing a blue hat, Code Review Engineer with pink hair, and Project Brief Manager with long orange hair. Each has a task number and status dot underneath. It looks like a team board, and also like a wall of employee badges.

Several interesting reactions appeared in the reply thread.

Someone said that the blinking design would make “firing an agent” a little weird. Yuanlin Lin replied: archiving one already feels a bit like a performance review.

Another person said their team ended up making an org chart, and the head of HR is an agent. Yuanlin Lin said that was crazy and that they should build it in Nuphos. Then he asked a harder question: does this org chart include humans? As CEO, for example, some of his human employees report to a “manager agent.”

Someone else asked a detail question: if a session is restarted, does the same face stay with the role, or does it identify that particular run? These two interpretations would create very different expectations. Someone also wanted a standardized icon format so they could put agent faces elsewhere.

Will raised a challenge from another direction. He said he didn’t understand why agents should be the top-level thing rather than tasks. Making the agent an entity makes it look like it can only do one thing at a time.

Yuanlin Lin admitted that this part wasn’t thought through. What they did was simple: rename “session” to “agent,” and give each agent a face. That meant assigning new tasks to old sessions, rather than letting an existing agent start a new session. In a reply, he asked: should the concepts be split apart so that an agent can own multiple sessions?

Naming changes usage. Call it a session, and users treat it as a record—use it and move on. Call it an agent and give it a blinking face, and users start managing it the way they manage employees: giving it a job title, looking at the org chart, doing performance reviews. Even “archiving” takes on an emotional charge.

Nuphos itself is an agent workspace for teams. You keep using Claude Code or Codex, and log in to the runtime with your own account. Sessions aren’t tied to a single machine; close your laptop and the turn keeps running, and you can keep watching on your phone. A session can invite colleagues in like a Notion page, with several people talking to the same agent and seeing the same set of tool calls. Skills and tools are shared once across the team, and everyone’s agents can use them. Slack and Discord can pull the same agent into a channel. Triggers let cron or webhooks start a conversation on their own.

These features solve collaboration and continuity problems. But users using it to “hire” was not part of the original design.

Giving AI a face is a very small product decision. But it turns a “session” into a “role,” and “history” into “employment history.” Users won’t stay at the abstraction layer you give them; they’ll fill in the gaps with the organizational experience they know best. Once words like hiring, reporting, performance, and departure are activated, the product has to face a whole set of expectations from human organizations.

The three avatars in the video are just decoration for now. By the time users really start asking “who is this agent’s manager,” “who owns its memory after it leaves,” and “which layer do humans sit at in the org chart,” the product will no longer be answering UI questions.

The GitHub repo is here: https://github.com/nuphos/nuphos