Lore

Human-Attribution/Accountability Pattern

Linear's agent defaults to attributing the work it creates — a ticket, a PR — to a responsible human owner, keeping agent-originated work inside normal human accountability structures. The exception is fully system-triggered work, such as an agent that detects and resolves a monitoring alarm end-to-end with no human request in the loop; there, the agent can own the item itself since no human judgment initiated it. This keeps agent output discoverable and owned without forcing every automated action through a human proxy. Distinct from but related to Agent Output Must Be Filed Into the System of Record: that principle is about where agent output lands (right project/status), this one is about whose name it lands under.

Self-Triggering Without a Human

Self-triggering without a human in the loop

The system can run fully autonomously with no human trigger at all: a monitoring alarm can cause the agent to file a bug against itself and attempt a fix, with no person initiating or approving the loop. This shows the "agent work is attributed to a human" framing is not a fixed policy but a spectrum — most work is nominally attributed to a requesting human, but the edges (alarm-triggered, self-initiated work) fall outside that framing entirely.

Contradictions

Earlier material frames human attribution as a stated rule for how agent work is accounted for; this part shows fully unattended, non-human-triggered agent action (monitoring-alarm-initiated bug filing/fixing) that the rule doesn't cover, meaning it functions as a default/spectrum rather than an absolute policy.