AI work attribution

WhoWorked records human hours and AI agent contributions in one delivery record. The person remains the headline. The agent activity sits beneath that work as evidence of how the result was produced.

What an attribution can contain

Depending on the reporting source, WhoWorked can record:

• The person connected to the work
• Source tool and model
• Start and end time or duration
• Project and related delivery context
• A summary of what was accomplished
• Token usage and cost when the source provides them
• Capture and reporter provenance

The available fields vary by tool. Missing token or cost data does not mean the contribution is invalid, but it changes which comparisons are safe to make.

Connect a reporting source

Choose the connection that matches where AI work happens:

• Use the hosted MCP server when an AI client should read or update the workspace directly.
• Install the browser extension for supported browser-side activity.
• Install the WhoWorked reporter where local AI tools produce attribution events.
• Use the API when your own system already has reliable contribution evidence.

Members with access to AI attribution can review reporter coverage and recent evidence under AI reporting. Owners and admins can also review workspace-wide connection settings.

Review attribution evidence

Open AI reporting to inspect member coverage, reporter health, and recent evidence. Open the attribution review area when records need a human decision before they become part of the trusted delivery view.

During review, check four things:

1. Person: Is the contribution attached to the person responsible for the result?
2. Project: Does the commercial and delivery context match?
3. Summary: Can a reviewer understand what was accomplished?
4. Provenance: Is the source expected and sufficiently complete?

Do not infer precise human time saved from token count alone. Equivalent output and leverage metrics depend on configured methodology and complete underlying records.

Keep the human as the headline

Attribution should explain amplification, not grade a person by the volume of their AI activity. Review output in the context of the person, project, and deliverable. A high token count can represent useful exploration, repeated failure, or a large input. The surrounding work tells you which.

Fix common gaps

A member has no recent evidence. Confirm the reporter or extension is installed, active, and connected to the same identity used in the workspace.

The project is missing. An attribution receives project context through its linked time entry; there is no direct project field on the attribution. Link it to an existing entry with the correct project, or create or edit the time entry first and then link the attribution.

The summary is vague. Replace it with a factual result. Avoid pasting prompts, source code, or private client material into the summary.

Evidence is duplicated. Keep one trusted record and investigate whether two reporters are observing the same source activity.

After evidence is complete, use reports and insights to read contribution alongside human time.

AI attribution review and AI reporting require the ai_attributions_enabled workspace entitlement.