Case Study — Agentic AI on Snowflake Cortex Code (Coco)

A tiered agentic framework for healthcare data operations

A large health plan engaged ArtiSoft to design an agentic framework on Snowflake Coco that monitors and maintains its healthcare data pipelines — cognitive judgment handled by agents, deterministic execution and final authority kept with people.

T0–T4
Autonomy tiers
0
Irreversible actions in the tool catalog
2-person
Approval on production transform code
The operational challenge

Data operations depended on scarce senior engineering attention

Claims, clinical, eligibility, and reference feeds move through dozens of pipelines. Monitoring and maintaining them fell almost entirely to a small group of senior engineers.

Alert fatigue

On-call engineers triage hundreds of threshold alerts a week; most are benign variation.

Silent schema drift

Upstream feeds change shape without notice, breaking or quietly corrupting downstream transforms.

Slow root-cause analysis

Every incident means correlating logs, lineage, and tickets across disconnected tools.

Maintenance vs. roadmap

Pipeline fixes and new development compete for the same engineers, so both backlogs grow.

Why this architecture fits

Cognitive judgment to agents, deterministic execution stays put

A canonical domain model is established before any agent is built, so agents emit structured, verifiable objects rather than plausible-sounding prose. Every process step is then classified as cognitive or deterministic — deterministic steps stay with conventional systems, cognitive steps go to agents, and every transition into execution requires human supervision. JetStore is the system of record for that boundary: it owns the domain model and executes every deterministic task, so agents are never asked to do a compute platform's job.

01
Domain model lives in JetStore
Source, Feed, Pipeline, Run, Anomaly, Incident, and Remediation are modeled once in JetStore. JetStore is the single canonical definition every agent, schema, and prompt derives from — no team maintains its own shadow copy.
02
Model compiles to vocabulary and schemas
JetStore compiles the domain model into the controlled vocabulary agents reason with and the JSON schemas that constrain structured output — the same entities and relationships, mechanically kept in sync rather than hand-copied into prompts.
03
Deterministic execution stays in JetStore
Joins, transforms, lookups, and rule evaluation run as deterministic JetStore compute — not agent inference — so anything with one correct answer gets one, reproducibly, every time.
04
Monitoring agents advise
Observer, Triage, RCA, and Communications agents detect anomalies, correlate evidence, and draft incident narratives — reading and writing only in terms the domain model defines.
05
Maintenance agents propose, engineers author
Schema Drift, Transformation Engineer, Test Generation, and Impact Analysis agents draft changes; a Reviewer agent adds independent review before a human approves.
06
Runs entirely on Snowflake Coco
Agent runtime, tool catalog, JetStore, and least-privilege roles live inside Cortex Code, so platform data never has to move to reach the agents operating on it.
Human supervision & control

Autonomy is assigned by tier, not assumed

Every capability is bound to a tier at design time, enforced by which tools are present in an agent's catalog. Tiers advance only against measured evidence, and promotion is automatically reversible if outcomes turn adverse.

TierAgent mayHuman role
T0 — ObserveRead metadata and code; record analysis to dashboardsNone; aggregate review only
T1 — AdviseProduce advisory hypotheses and incident narrativesConsumer of advice; final decision-maker
T2 — ProposeGenerate staged, non-executing artifacts (diffs, plans)Reviewer and author-of-record
T3 — Act with approvalExecute specific actions after authorizationApprover for every instance
T4 — Bounded autonomyExecute reversible actions within an authorized envelopePost-hoc reviewer; envelope owner
Savings path & velocity

Time back for engineers, earned in phases

Observer and Triage agents cut time spent correlating evidence during incidents. Schema Drift and Transformation Engineer agents turn maintenance from ground-up authoring into review. Reduced firefighting frees senior time for roadmap work.

35–45%
Faster triageFrom automated evidence correlation
2–3x
Faster drift-to-fixAgents draft, engineers review
~30%
Hours redirectedTo roadmap work by Phase 2
12–18mo
Typical paybackFollowing a Phase 0 baseline
Illustrative projections, not commitments and not measurements — the shape of the business case rather than its final numbers. They firm up once Phase 0 captures the organization's current MTTR, backlog age and reviewer throughput, which is the proposal's own method: benefit is claimed only where a baseline exists to measure it against.
Guardrails

Enforced at the tool and platform layer

Designed into the framework rather than left to prompt instructions. Each is enforced where it cannot be talked out of — in the tool catalogue, the platform roles, or the audit path.

Context-aware privacy

PHI enters agent context only when essential, preferring de-identified or aggregate representations.

Write-before-act audit

If a write to the audit store fails, the associated action does not happen.

Two-person approval

On production transformation code; the approver may not be the engineer who initiated the task.

Input isolation

External feed content is treated as non-instructional, so no feed can invoke an agent capability.

Kill switch

A single control suspends all agent execution while the underlying platform keeps running.

No self-modification

Agents cannot alter their own prompts, permissions, or guardrails.

See the domain model and tier assignments for your pipelines.

We'll walk through the domain model, the monitoring and maintenance agent designs, and a Phase 0 plan sized to your data ecosystem on Snowflake Coco.

Request the Full Proposal Email me the case study

This engagement was designed for the client's platform. ArtiSoft's own recommendation is JetStore Agentic AI, where the reasoning layer runs on inference you host and no external LLM provider is required — the platform accommodates either.

Contact

Have a question? Let's talk.

Email us to discuss the domain model, autonomy tiers, or a Phase 0 scope for your data platform.

info@artisoft.io