Most AI agent products treat the org chart as a settings page. We treat it as the main interface. Here is the design argument for why, and what it shaped.
14-day trial. No DevOps. No Sales call. Provisioned in under a minute.
The Agent Hive home screen is an org chart. You log in, you see a canvas of cards arranged in a hierarchy: the CEO agent at the top, department leads underneath, agents underneath them. You can drag cards to reassign. You can rename them. You can right-click to fire one. You can double-click to open a side drawer with the agent's job description, recent work, and budget.
This is not an obvious choice. The default shape of an agent product is a chat thread, or a job queue, or a list of "skills" with toggles next to them. We went the other direction on purpose. This post is the argument for why.
In Agent Hive, an agent is not a model call. An agent is a role.
A role has a job description, a set of tools it is allowed to use, a budget, a reporting line, and a human approver for risky actions. Two agents with the same model and the same prompt are different agents if their tool inventories or reporting lines differ. The agent is the thing you can hire, fire, promote, give a raise (a higher budget), or move under a different lead.
Once you accept that an agent is a role, the org chart is not metadata. The org chart is the agent layer. The CEO agent's reporting line determines which approvals it sees. A marketing-ops agent reporting under the marketing department head sees that department's brand tone and gets escalation routing to that lead. Move the same agent under sales engineering, and its context, its approvals, and its budget envelope all change.
If the org chart is what determines how agents work, putting it in a settings page is the wrong call. Settings pages are the place you visit once during onboarding and then forget. The org chart is something you should be able to touch every day, the way a human leader of a small team rearranges who reports to whom as the work changes.
We built two interfaces before this one. They are worth describing because the org chart only makes sense in contrast.
Version 1 was a job queue. The home screen was a table of in-flight work. Each row was a task an agent was doing or had finished. You could click into a row to see the steps, the cost, and the result. This worked for one agent. With two agents it started to feel cramped; you wanted to know whose work each row was, and the column got busy. With four agents, the table was unreadable, and we found ourselves filtering by agent so often that the filter dropdown was the most-used UI element on the page. The thing the filter dropdown was a proxy for was an org chart.
Version 2 was a chat thread. The home screen was a conversation with the CEO agent. You typed what you wanted, the CEO agent decided which sub-agent to route to, and the result came back inline. This was a much better demo than version 1, and it was actually worse to operate. The problem was that everything happened in one stream. You had to scroll to find out what a specific agent was doing right now, and the moment two agents were working in parallel, the chat thread became unreadable for the same reason a Slack channel with two parallel conversations is unreadable.
The org chart fixes both. Each agent is a card. Each card has a presence indicator that shows whether it is idle, working, or waiting on approval. You see at a glance who is doing what. You don't need to filter. You don't need to scroll.
A few specific things shipped that would not have made sense in a settings-page treatment.
None of those operations make sense as a tab inside a settings page. All of them make sense as a canvas operation.
The org-chart-as-product choice closes some doors we are sometimes asked about.
If you have a strong opinion about what the org chart should let you do that it currently does not, the contact form goes straight to me. The org chart is the interface I expect to spend the most calendar time on through 2026, and it is the surface most likely to look different in six months.