polite.aiDocs
BUILD / Agents & teams

Teams, transfers & subagents

Agents work together two ways — transfers move the caller, subagents help quietly in the background. Good teams use both.

Teams let each agent be one thing done well. The team graph draws the two kinds of collaboration distinctly, because they feel completely different to a caller.

Transfers: the caller moves

A transfer hands the live call to another voice agent — reception to bookings, English to the caller's own language. The receiving agent gets a handover summary, so the caller never repeats themselves, and each agent keeps its own voice — the caller hears a new colleague pick up, with zero hold time.

Transfers to humans

Agents also transfer callers to real phones, two ways:

  • Blind — "let me put you through", and the call simply goes.
  • Warm (consultative) — the agent rings the colleague first and briefs them privately on who is calling and why; the human accepts or declines without the caller hearing it. If the human declines, the reason comes back so the agent can pivot — take a message, book a callback, or try the next person.

Out-of-hours handling is part of the agent's instructions — transfer during office hours, capture a message outside them; given the hours, the Builder writes the switch in. Every transfer destination passes the outbound safety filter — see What every agent promises your callers.

Hand-back: the human talks, the agent does the paperwork

After a transfer on a phone call, the colleague can key a short sequence on their handset to hand the caller back to an agent. A transcript of the human side of the call briefs that agent — transcription is best-effort, so the agent confirms what was agreed in one sentence — and the agent finishes the admin: books the slot, captures the details, emails the summary.

The Human hand-back section of the agent editor maps each key sequence to the agent that picks the caller up; the transcript briefing can be switched off there. Hand-back works on both blind and warm transfers, on calls that arrived by phone — not on browser calls.

Subagents: quiet helpers

A subagent is an agent used like a tool: the front agent asks it a question mid-call and gets the result back, while the caller stays exactly where they are. Text agents shine here — a knowledge specialist answering from a knowledge base, an email drafter, a validator. A subagent call takes a good few seconds, so well-built agents say "let me look into that…" first — the Builder writes that in.

Designing a reception-and-specialists team

The classic shape: one receptionist answers, greets and routes, with a specialist behind it for each real job — each with its own voice, prompt and tools — plus a shared knowledge subagent any of them can consult. Given a description of the departments, the Builder proposes exactly this; the team can then be tested end to end before going live.

Good to know

Which agents can transfer, hand back or call subagents depends on the agent type — the Builder only proposes what the design can actually do. See Models & platforms.

Last updated 2026-08-17