Understand Token Efficiency

Token Efficiency helps employees and Owners understand how much agent model usage is associated with observed output. It is a directional signal for investigating workflows, not a score of employee quality or effort.

Gesta measures output when the agent writes it. It does not estimate authored work from the final file size, commit, or Git diff, where formatters and generators can obscure what the agent actually produced. As a result, Gesta output measurements are not expected to match repository line counts exactly.

Personal and team views

Personal efficiency appears inside My Activity, alongside personal usage, produced output, model trends, and work-type mix. Owners use Observability → Token Efficiency for the organization summary and member comparison.

Use these views to understand:

• which workflows use substantial model capacity;
• where long sessions produce little observable output;
• whether model or workflow changes affect usage;
• where additional guidance or training may help.

Interpret the metric carefully

The same token usage can represent very different work. Investigation, debugging, incident response, and verification may require more context than a small source edit.

Before acting on a trend, ask:

• What type of work was being performed?
• Was the final output captured by Gesta?
• Did the session include investigation or verification?
• Did a failed approach lead to a later successful result?
• Did model pricing or model selection change?

A lower number is not automatically better. The goal is useful, reliable work with proportionate model usage.

Improve a workflow

Use Token Efficiency to identify a workflow for review, then inspect a small sample of sessions:

1. check whether prompts state the goal and constraints clearly;
2. provide relevant organization context;
3. remove repeated or contradictory instructions;
4. verify that the agent is using the right tools;
5. compare results after the workflow changes.

Do not optimize by removing necessary testing, safety checks, or evidence.

Model pricing

Cost estimates depend on the pricing catalog configured by the organization. Owners should keep model input and output rates current and use the correct effective date. See Organization settings.

For session-level context, use the Observability overview.