What Scaled
Internal-facing, narrow-scope, high-volume agents with clear escalation. Password resets, case summaries, knowledge drafting, meeting prep. Users save time; risk is low; adoption compounds.
What Stalled
Broad customer-facing agents without rigorous evaluation. Autonomous agents in high-stakes workflows. Agents deployed without observability — problems compound invisibly.
Common Failure Modes
Prompt hygiene neglect (prompts drift and no one notices). Trust erosion from un-audited changes. Cost surprise from unmetered rollouts. Integration fragility (MCP/spoke failures cascade).
Generalizable Principles
Start narrow. Instrument thoroughly. Measure outcomes before scaling. Audit everything customer-facing. Human-in-loop for high-stakes; autonomous for deflectable. Phase trust carefully.