Noetive designs the sensing pod that feeds its self-improving factory model

Somebody at the buyer's plant has to own the pod: where it sits, and who deals with it when it stops reporting. Noetive designs that multimodal sensing pod itself, and it is the route by which physical conditions reach the model ^[7]. For an operations manager, that turns a software pilot into a hardware install.

Amir Frenkel, the chief executive, said most AI products to date "were designed for information work living on the internet" ^[3]. Noetive's answer is what it calls an "intelligence of record", built on a self-improving model the company refers to as a brain ^[4]. The agents on top of it run on the tools a customer already has ^[5]. That leaves two layers producing plans. When the agent's plan and the plan in the incumbent system disagree, the plant needs a rule about which one the shift follows, and the customer is the one who has to write it.

The one outcome in the launch announcement comes from Steuben Foods, whose chief executive, Menachem Katz, said work "that once happened monthly and took a week of planning now happens daily and takes minutes" ^[9]. Read the week as five working days in a month of about 21, and planning was taking close to a quarter of the calendar, while the new cadence means roughly 21 runs a month in place of one ^[18]. Katz said planning has always been one of the most time-consuming parts of running the business because even small changes ripple across the operation ^[10]. Noetive reads both Steuben's system data and conditions on the factory floor ^[8].

Eclipse Ventures led the round, and Noetive took shape inside the firm, which helped recruit the founding team ^[12]. Bloomberg reported in April that Eclipse had hired Frenkel from Meta as its first chief AI officer ^[13]. He was a Meta vice president for nearly a decade and has held leadership roles at Alphabet and Amazon ^[11].

The claim to press on is the improvement loop: each deployment is supposed to make the model more capable ^[6]. SiliconANGLE did not report a partner count, a price for the pod, or whether the software runs without it ^[19]. Before signing, I would want the contract to say whether one plant's data trains a model that another customer runs against.

Two questions separate a pilot from a construction project here. First, does the number move on system data alone? Field testing happens with a curated group of design partners ^[14], so an early buyer has standing to ask for one phase without the pod and one with it. Second, what does the shared model retain from your floor? If nothing moves until the pod is mounted, the purchase is sensing hardware with software attached. In a pilot like this I would track planner hours and the share of daily plans that get executed unchanged; a count of agent sessions would not tell me whether the plant runs better.