Feature Creature Sightings — September 15, 2026

This run has two current sightings, but neither justifies a victory lap. One is a useful diagnostic for enterprise email systems; the other is an AI model announcement whose operational value remains unproven.

Enterprise email problems are usually systems problems (note)

Crittr verdict: ⚖️ Situational

A newly published vendor guide examines what starts breaking when email programs and contact databases move beyond ordinary campaign scale. Its useful territory is infrastructure, governance, and measurement—not another list of subject-line tricks.

Larger databases make weak data controls, inconsistent measurement, and inadequate sending infrastructure harder to ignore. Those failures affect deliverability, personalization, and attribution, so teams evaluating enterprise email tooling need to assess the surrounding CRM architecture rather than compare campaign features alone.

Technicole take: Treat this as vendor-authored reference material, not independent research or a new market development. The sound takeaway is that contact volume exposes architecture debt: buying an upmarket feature tier will not resolve unclear governance, unreliable data, or attribution definitions that teams never agreed on.

Source: HubSpot Marketing Blog

Koa is a model launch, not yet a CRM operating model (note)

Crittr verdict: 🪲 Feature Creature

Salesforce launched Koa, a reasoning model built on Nvidia's Nemotron framework and aimed at sales, marketing, and customer-support use cases. The announcement establishes a new model option, but the supplied material does not show how it performs inside real CRM processes.

A domain-oriented model could become relevant if it handles the context, permissions, and decision paths involved in customer-facing work. A launch alone does not demonstrate that it can quote correctly, update records safely, preserve handoff context, or produce decisions teams can audit.

Technicole take: Ignore the competitive-threat framing for now. The operator question is not whether one AI lab should fear another; it is whether Koa can execute bounded CRM tasks more reliably than existing options without creating new review queues and data-quality problems. There is not enough evidence here to answer that.

Source: TechCrunch (AI)

The common trap is mistaking a purchased capability for a repaired system. Enterprise email features and reasoning models both inherit the data definitions, controls, and ownership gaps already present underneath them.