Brief bench
Tests whether the task contains enough firmographic, technographic, and regional context to produce a useful shortlist.
AN AGENT-NATIVE EXPERIMENT
11x AI is presented here as an operating model: translate a sales question into structured research, preserve uncertainty, and let a person decide what happens next.
“The useful unit of automation is not message volume. It is a decision-ready result whose inputs, limits, and next action are visible.”
This principle shapes the experience: a task begins with an explicit ideal customer profile, moves through configurable data access, and stops at review gates before outreach. The agent can accelerate repetitive research without pretending that a title proves authority, an intent signal proves demand, or an email match guarantees deliverability.
EVALUATION FACILITIES
Tests whether the task contains enough firmographic, technographic, and regional context to produce a useful shortlist.
Surfaces source, timestamp, verification state, and data-provider dependence so evaluators can challenge a result.
Holds targeting and message decisions for a person rather than treating faster automation as permission to send.
Prompts operators to read installation output, restrict credentials, test revocation, and document retention behavior.
FIELD NOTES
Record which providers contribute each field, how conflicts resolve, what a miss means, and whether additional coverage justifies added governance.
Define the observed event, time window, account matching logic, and corroboration requirement. Treat prioritization as a hypothesis, never a purchase guarantee.
Document who approves the list, the claim, the message, and the send. SPF, DKIM, DMARC, suppression, consent, and opt-out controls remain operational responsibilities.
Test update and removal steps, revoke provider keys separately, and verify deletion across connected systems instead of assuming package removal clears remote data.
Copy the command, review what it will execute, then run a deliberately narrow first task.