All docs

Agent system · 10 min

601 named agents. Six departments. Five tiers.

The KAIRO workforce is a 4-tier hierarchical system led by a single CEO agent. Every agent has a name, a model, a role, and an accountability mode. The accountability rules are encoded into the platform code, not stored in a policy document.

The hierarchy

Six departments. Five tiers. 601 named agents. The split:

  • Tier 1 · CEO Atlas (1 agent). Premium reasoning model for strategic work, standard model for routine. Routes every workflow.
  • Tier 2 · Department Managers (6 agents). Brennan (SDR), Marcus (Marketing), Lena (Research), Riley (CS), Vance (Ops), Sterling (Executive).
  • Tier 3 · Team Leads (30 agents). Five per department.
  • Tier 4 · Senior Specialists (300 agents). Fifty per department. Activated on demand.
  • Tier 5 · Workers (264 agents). Pure execution. Cost-optimized worker models for text and fast retrieval models for live search.

The accountability modes

Every agent operates under one of three modes:

Human-gated · 231 agents

Output cannot ship without explicit human approval. This covers every customer-facing surface (emails, posts, comments), every financial decision (pricing, comps, refunds), and every irreversible action (deletes, publishes, sends).

Human-notified · 82 agents

Output ships, but you are notified for review. This covers internal research findings, flagged account health changes, and internal scoring or triage events.

Auto-execute · 288 agents

Internal, no human approval needed. Data hygiene, dedup, internal reporting, memory storage, draft generation (drafts only, never sends).

Heads upDrafting is auto-execute. Sending is human-gated. KAIRO drafts everything, but ships nothing customer-facing without you. The split is hard-coded.

The 5-agent operating layer

Every tool run assembles a 5-agent team:

  • 1 Manager who owns the directive.
  • 1 Lead who runs the work.
  • 3 Specialists who execute, critique, and harden the output.

The team selection is deterministic per tool. The same tool with the same input gets the same team, every time. That makes the platform debuggable.

Delegation in practice

You issue a directive in the control room. Atlas routes it to the right department manager. The manager spawns named leads. The leads spawn specialists. Each step is logged. The full trail is visible on the delegation console at /admin/delegate for admin users.

A live delegation tree for a typical churn directive runs 6 messages and costs approximately $0.007 end to end. The platform is designed for that kind of cost discipline.

Why this matters

Most AI platforms have a flat agent pool. KAIRO has a workforce. The hierarchy is not decoration. It is what lets you delegate a strategic directive, watch it cascade through the right specialists, and review a hardened output without burning a senior model on a worker task.