Skip to main content

​
Purpose

Two things are true at once, and most vendors pick one. Horizontal platforms win on reuse. Governance, identity, approvals, audit, integration and evaluation are the same problem in every business function, and solving them once is worth far more than solving them five times. Vertical applications win on depth. A generic agent platform hands a customer a toolkit and leaves them to invent the entire evidence model for their function. In a regulated domain that is not a head start; it is the whole project. Smaya is built on the position that you need both, in that order.

​
The architecture

Exterview runs on Smaya today. Dotted lines are Upcoming: Gradientflo and later workloads on the same platform.

​
Land with Exterview. Expand with Smaya.

An organization does not buy a platform. It buys a solution to a problem it already has.
1

Land

A concrete problem — usually hiring quality, hiring speed, or the inability to defend a hiring decision. Exterview solves it, and the customer gets a governed digital workforce doing real work.
2

Prove

The controls prove themselves on live decisions: bounded authority, human approval, a complete and reproducible record. Security and risk see the model working before it is extended.
3

Expand

The governance, integrations, identity and evidence model are already in place and already approved. The next workload is a configuration and commercial step, not a second implementation.
That sequence is the whole strategy. The hardest thing to earn in an enterprise is permission for AI to operate on consequential data. Once earned, extending it is comparatively easy — which is why the platform matters and why the first application has to be genuinely excellent rather than a demo.

​
Why depth is defensible

The general-purpose agent platforms will out-build anyone on tooling. That is not where the defensibility is. No horizontal platform will ship the machinery of a regulated employment decision — continuous adverse-impact measurement, an evidence trail that ties a judgment to the criteria and the material that supported it, notice and human-review rights, and reproducibility that survives a challenge years later. Their buyer does not need it, and their category does not carry the liability. We do, and we designed for it from the start. Evidence trails, score provenance, notice before every AI interview and human-only hiring decisions are in the product today; continuous adverse-impact measurement is Upcoming. That work then becomes reusable across every function where decisions about people carry weight.

​
What we deliberately do not build

Not an open agent marketplace

Every agent in the library is one we built, tested and stand behind. An agent making judgments about people carries obligations that cannot be delegated to an unknown publisher.

Not a system of record

Your HRIS holds the employee. We hold the judgment and the evidence behind it. If the output is a record, it is not ours.

Not a blank development canvas

Customers configure pre-built agents rather than authoring them. The evidence standard behind an agent is the product, and it cannot be assembled in an afternoon.

Not autonomous decision-making

Consequential outcomes are decided by named humans. The platform prepares the evidence. That is a design commitment, not a current limitation.

​
FAQs

It is a horizontal Agent OS with vertical applications on top. The platform capabilities are general; what we ship on them is deliberately domain-specific, because that is where the value and the liability both live.
Nothing stops them building the platform layer — they will. What they will not build is the regulated-decision machinery for one function, because their buyer does not require it. Depth in a liability-bearing domain is the durable position.
The library grows toward 100 agents across the platform, sequenced by customer demand. See Roadmap to 100.