Verticals
Where specialist models can beat frontier AI.
Agentsia is for domains where Claude, ChatGPT, and Gemini are the benchmark, not the operating model. The fit appears when a buyer has a hard latency budget, a strict data boundary, a repeatable workflow, and a governance team that needs evidence before promotion.
Today, the public evidence is adtech-first: Assay-Adtech v1 is the active released benchmark surface. Fintech, health, automotive, and on-device remain future-domain scaffolding until each has its own frozen rubric, dataset, and results.
Latency pressure
Data boundary
Governance proof
Deployment topology
Constraint matrix
A vertical is a fit when constraints rule out a remote general model.
The decision must clear before the bid expires.
Active evidenceThe model must explain a decision without exporting customer financial data.
Future candidateThe workflow must follow local protocols while patient data remains inside the approved boundary.
Future candidateThe model must operate when the network is slow, absent, or irrelevant to safety timing.
Future candidateThe useful model must fit memory, battery, and local privacy constraints.
Future candidateBounded workflow, controlled boundary, measured promotion.
Remote latency, data egress, or weak evidence posture.
The matrix separates the active adtech evidence from future domains. Non-adtech rows show where the same control-plane pattern may apply, not a claim of shipped proof.
See the three-layer stackPositioning
Modelsmith is below the agent app, above the inference substrate.
The buyer can still use frontier-lab models as baselines and agent platforms as application surfaces. Modelsmith earns its place when the model itself needs a governed operating loop and a customer-controlled deployment path.
Dimension
Frontier-lab APIs
Generic AI platforms
Modelsmith
Frontier-lab APIs
Remote model endpoint owned by the frontier lab.
Generic AI platforms
Hosted workflow tooling with model choice delegated to providers.
Modelsmith
Designed for customer-controlled specialist artefacts, runtime, and boundary.
Frontier-lab APIs
Public benchmark strength, weak fit to private workflow gates.
Generic AI platforms
Task automation metrics, often detached from model promotion.
Modelsmith
Governed evals with promotion evidence, rollback, and lineage where the operating loop is enabled.
Frontier-lab APIs
Strong for analyst work, structurally wrong for tight paths.
Generic AI platforms
Useful for back-office orchestration, not real-time control.
Modelsmith
Small specialists run where the decision happens.
Frontier-lab APIs
Sensitive payloads can leave the customer environment.
Generic AI platforms
Connectors expand reach, but also expand data movement.
Modelsmith
Designed so training data, rubrics, weights, and evidence stay controlled.
Frontier-lab APIs
Vendor release cadence and API behaviour drive change.
Generic AI platforms
Workflow changes ship faster than model governance.
Modelsmith
Designed to connect eval, post-training, promotion, rollback, and runtime targets into one loop.
Frontier-lab APIs
Per-token or per-call cost scales with every decision.
Generic AI platforms
Seat and platform fees before model ownership.
Modelsmith
Fixed controlled infrastructure can suit high-volume bounded work.
Buyer fit
One active vertical, four future fit hypotheses.
The page is not a list of active launches. Adtech is the current evidence base. The other verticals stay visible because their buyer constraints shape the same control-plane architecture, but they remain future candidates until their own evidence exists.
Vertical
Buyer pressure
Why a specialist
Evidence
Topology
Ad tech
DSP, SSP, gaming monetisation, or retail-media team
The decision must clear before the bid expires.
A small specialist runs next to the bidder, scores only the bounded auction task, and promotes only after held-out RTB scenarios clear.
Current public proof is adtech-first: Assay-Adtech v1 is the active released benchmark surface.
Co-located training and inference
Fintech
Risk, fraud, credit, or compliance team
The model must explain a decision without exporting customer financial data.
A specialist could learn institution-specific fraud and policy patterns inside the customer environment once the fintech evidence path is intentionally activated.
Fintech remains scaffolding until a customer-approved rubric and public Assay release are available.
Trained centrally, deployed to a commercial runtime
Health
Hospital, trust, payer, or clinical-software team
The workflow must follow local protocols while patient data remains inside the approved boundary.
A specialist would need approved clinical note, pathway, and discharge examples, then human clinical gates, before this vertical becomes active.
Health remains scaffolding until clinical reviewers, rubrics, and safety gates support a public evidence claim.
Trained centrally, deployed to a commercial runtime
Automotive
ADAS, autonomy, connected-vehicle, or service engineering team
The model must operate when the network is slow, absent, or irrelevant to safety timing.
A specialist would need to be trained centrally, signed, and deployed to the vehicle or service runtime with evidence tied to the model version.
Automotive remains roadmap scaffolding until evaluator design and review evidence are available.
Trained centrally, deployed to the edge
On-device
Mobile, desktop, embedded, or privacy-first product team
The useful model must fit memory, battery, and local privacy constraints.
A sub-4B specialist could be trained for one narrow product workflow and exported to the customer-selected edge target once the edge-fit evidence path is active.
On-device remains candidate scaffolding, adjacent to adtech gaming-SDK evidence but not a released vertical of its own.
Trained centrally, deployed to the edge