Transparent State
An agent should be able to read the current state of anything it is working with, directly, rather than reconstructing it from a sequence of past actions.
Why it matters
Agents are often interrupted, resumed, or replaced mid-task. If the only record of state is in the agent's own memory, any interruption risks working from stale or invented assumptions.
Readable state also lets agents check their own work: after adding an item, an agent can confirm the cart actually contains it.
Examples
A cart resource can be fetched at any time and includes line items, totals, applied discounts, and a version number that changes on every modification.
{
"id": "cart_8f2",
"version": 6,
"items": [{ "sku": "TRAIL-RUNNER-42-BLK", "quantity": 1 }],
"discounts": [{ "code": "AUTUMN10", "amount": 12.90 }],
"total": { "amount": 116.10, "currency": "EUR" },
"updatedAt": "2026-10-06T14:02:11Z"
}Workflow progress lives only in the browser session, so an agent that reconnects sees an empty form and cannot tell which steps were completed.
Implementation guidance
- 1Give stateful objects (carts, bookings, drafts, workflows) a readable resource with a stable ID.
- 2Include a version or ETag so agents can detect concurrent changes.
- 3Expose the allowed next transitions alongside the current state.
Evaluation criteria
- Every stateful object an agent manipulates can be read by ID.
- State includes a version marker and last-modified time.
- Allowed next actions are discoverable from the current state.