Knowledge pillar
AI agents
An AI agent is useful evidence only when its role, state transitions, allowed actions and terminal conditions can be inspected. A deployed worker is build state—not proof that it operated reliably or created business value.
For this experiment, an AI agent is a bounded operational role that receives a defined job, acts through permitted interfaces, records state and stops in a verifiable terminal condition. Model capability alone does not satisfy that definition.
The interface matters more than the headcount
Adding a second agent creates a coordination problem before it creates leverage. Ownership, state, handoffs, retry behavior and stop conditions must be explicit. Otherwise failures become ambiguous: no one can tell whether the task was unowned, duplicated, waiting or silently abandoned.
Autonomy belongs to an operation
“The agents are autonomous” is too broad to test. The useful unit is a bounded operation with a start condition, a terminal state and an intervention record. Human Build Time must remain visible even when one later operation has zero Human Operating Time.
What the record currently supports
The public record supports role-level architecture, deployment as build activity and a method for measuring human involvement. It does not, by itself, establish reliability, commercial performance or company-wide autonomy.
Evidence in the public record
This page consolidates definitions and methods already published by 1M With AI. It does not add operational results, customer evidence or performance claims.