Solved for drafting. Summarizing, surfacing, briefing, pre-call. The productivity layer a platform ships — and the ceiling it hits.
Not solved for prediction. Churn forecasting fails on context. The agent that reads the account weekly and hands the CSM a call they can defend.
Code decides, model narrates. The risk math stays deterministic and auditable; the LLM only writes, grounded and human-gated.
Trust is the moat. Shipped evidence, not claims. The agent never sends; a human does. Outcome measured in revenue, not hours.
The fix is four moves, and the agent is all four.