Linear's Delegation Model for Human-Agent Accountability
TRIGGER
When agents were assigned issues like human teammates, accountability became murky—agents accumulated dozens of assigned issues with no clear human owner. If someone disagreed with an agent's actions, there was no obvious person to escalate to. The standard assignment model implied transferred responsibility, but agents can't actually be held accountable for outcomes.
APPROACH
Linear introduced 'delegation' as distinct from assignment. Issues can only be assigned to humans and only delegated to agents. Input: User action to involve an agent in an issue. Output: Issue record with both a human assignee (accountable for outcome) and a delegated agent (responsible for taking action). The UI always shows both, making the accountability chain visible. Sub-issues were considered but rejected because they're intentionally independent with their own status and owner.
PATTERN
“When your agent accumulates 50 assigned tasks, you'll hear 'who's responsible for this mess?'—and have no answer. Model agent involvement as delegation (agent acts, human accountable) rather than assignment (responsibility transfers) to make the accountability chain explicit in your data model.”
✓ WORKS WHEN
- Agents take actions that require human review, approval, or correction
- Organization needs clear accountability chains for compliance, escalation, or performance tracking
- Multiple humans might interact with an agent's output and need to know who is responsible
- Agent work products have real business consequences where blame-assignment matters
✗ FAILS WHEN
- Agents handle fully autonomous tasks with no human review loop (automated cleanup, scheduled jobs)
- Accountability is tracked at the agent/system level rather than per-task (logs, monitoring)
- Single-user context where the user is obviously accountable for their own agent's actions
- Task model doesn't have assignment semantics to begin with