Skip to main content

AI Usage > Actor type

Written by Tom Williams

Dataset: AI Usage

Entity: AI Usage Measurement

Field ID: actor_type

Type: Select list

Description: The nature of the actor the measurement is attributed to. Possible values are:

  • HUMAN a person.

  • SERVICE_ACCOUNT an API key or service account.

  • AGENT an autonomous agent run.

Source: Calculated

Transformation logic: Calculated from the actor fields the provider populates on a record.

App Mapping

GitHub Copilot

Calculated: always HUMAN, Copilot reports per-seat usage only

Anthropic Claude (coming soon)

Calculated: HUMAN from actor.type = user_actor, SERVICE_ACCOUNT from actor.type = api_actor

Cursor (coming soon)

Calculated: AGENT where the event carries a cloud agent or automation, SERVICE_ACCOUNT where it carries a service account, HUMAN otherwise

OpenAI Platform (coming soon)

Calculated: HUMAN where a user owns the API key, SERVICE_ACCOUNT where only a key is identified

OpenAI Codex (coming soon)

Calculated: HUMAN on per-user rows

Reporting Use Cases

The Actor type field separates people from machines, which matters because agent and service-account usage can dwarf human usage in token terms.

  • Honest per-developer figures: Filter to HUMAN before dividing any measure by a headcount. Service account traffic would otherwise be attributed to whoever owns the key.

  • Measuring autonomous work: Filter to AGENT to isolate cloud agent and automation runs. The record still carries the person who owns the run in actor_username, so you can attribute autonomous work without a separate field.

  • Explaining token spikes: A sudden jump in tokens with flat interactions is usually agent or API traffic. Grouping by actor type shows it immediately.

Did this answer your question?