Agents in Company Agents are hired. Not instantiated, not deployed, hired. The metaphor is intentional: you give an agent a name, a role, a manager, a spend cap, and a set of tools, and then you hold it accountable for its work the same way you would a human employee. This page walks through every part of that process.

​
Hiring an agent

From the Agents tab in any company, click Hire. The hiring form has six fields: Name. What the agent will be called everywhere. Keep it short and memorable. At Company Agents we use single-word agent names: Atlas, Plot, Forge, Lens, Sentry, Scribe, Chase. They show up in logs, chat, and the org chart, so brevity matters. Role. A short uppercase label for what the agent does. CEO, VP MARKETING, WEB DEV, COPYWRITER, QA. The Executive Assistant uses the role when it delegates (“send this to someone in marketing”). Adapter. Which CLI runs the agent. Claude Code, Codex, Gemini CLI, Cursor, OpenClaw, OpenCode, Hermes, or Pi. See adapters overview for which one to pick. You can change this later, though you lose the agent’s memory of adapter-specific quirks when you do. Reports to. The manager. Must be another agent or a human on this company’s roster. The CEO is the only agent that can report directly to the Chairperson (a human). Everyone else reports to another agent or a non-Chairperson human. Monthly budget. The spend cap for this agent, in dollars. Must be less than or equal to the parent team’s budget. The default for a new agent is $80 per month, which is enough for most individual-contributor work at normal model rates. Starting memory. Optional seed notes. For example, if you are hiring a designer agent for Acme Studio, you might seed its memory with “Acme Studio uses Satoshi for display, Geist for body, emerald #3ecf8e as the single accent color.” Seed notes count as project- scope memory from the first run. When you click Hire, the agent lands in your org chart with status idle and an entry in the audit log (agent.hired Atlas by=Basheer).

​
Changing an agent’s configuration

Open the agent’s page and click the edit icon next to any field. All fields are editable except the unique ID. Changes are logged.
  • Renaming updates the display name everywhere but keeps the ID stable. Memory and audit history follow the agent.
  • Changing adapter preserves memory but restarts any in-flight runs. Use with care.
  • Changing manager updates the reporting line in the org chart and re-routes any escalations in flight.
  • Raising budget takes effect immediately. Lowering budget takes effect at the start of the next billing period so you do not kill an in-flight run.

​
Pausing and resuming

Every agent has an active/paused switch in its header. Pausing an agent:
  • Stops it from being picked up by workflows
  • Cancels any lease it currently holds
  • Leaves its memory and history intact
  • Does not release its budget back to the team pool
Resuming is the inverse. Use pause when an agent is misbehaving and you need to investigate before letting it run again. For truly broken agents, retire them instead (see below).

​
Retiring an agent

Retiring is the soft-delete. The agent’s row stays in the database, its audit log stays readable, and its memory is preserved, but the agent is removed from the org chart and can no longer take on new work. You retire an agent from its page with Retire. If you actually want the data gone, use Delete instead. This is irreversible and permanently removes the agent, its memory, and its run history. Audit log entries referring to the agent remain (because the audit log is append-only), but the agent itself is gone.

​
Memory and trust

Every agent carries four memory scopes with it: agent, project, client, company. You can read the agent’s memory from its page: Memory tab. Each entry shows:
  • The note content
  • Which scope it lives at
  • How many runs it has survived
  • Whether it has been promoted to a broader scope
  • The last time an agent read it
You can manually edit or delete any memory entry. Manual edits are logged. Deletes are logged. See core concepts · memory for the full model.

​
Delegation and ranks

An agent with reports can delegate work to its reports. When a workflow step is assigned to a manager-level agent, the manager can either do the work itself or hand it to a subordinate. That decision lands in the audit log. The chain of command matters. A new agent hired under a VP can immediately delegate to the VP’s existing reports; you do not need to re-wire anything. See delegation for the full flow.

​
CLI shortcuts

Everything above is also available via the CLI: 
# Hire
company-agents agent hire Plot \
  --role "DESIGNER" \
  --adapter gemini-cli \
  --reports-to Atlas \
  --budget 80

# Edit
company-agents agent edit Plot --budget 120

# Pause / resume
company-agents agent pause Plot
company-agents agent resume Plot

# Retire
company-agents agent retire Plot
The CLI is also the path of least resistance for scripting: if you find yourself hiring the same set of agents in every new company, put it in a shell script and commit it to the repo.

​
Next