06-reference/research

snowflake intelligence vs cortex ai boundary

2026-07-07·research-brief·source: deep-research
snowflakecortex-aisnowflake-intelligencephdataagentic-platform

Snowflake Intelligence vs Cortex AI: The Real Product Boundary (mid-2026)

The question

What is the actual product boundary and feature split between Snowflake Intelligence and Cortex AI as of mid-2026, and — for a phData Deal Solutions Architect — where does a CAF/Brigade deployment plug into that stack? phData's project index names BOTH as targets, but no vault doc maps their real feature boundaries, so every discovery conversation risks imprecise framing. The right placement matters for the phData-Snowflake pitch: the DSA has to say cleanly which capability lives where.

What we already know (from the vault)

What the web says

The boundary is real and consistent across sources — but the names changed at Snowflake Summit 2026 (June 2, 2026), which is the single most important thing for a mid-2026 conversation:

Convergences and contradictions

Synthesis for RDCO

The crisp boundary statement. Snowflake Intelligence (rebranded Snowflake CoWork at Summit 2026) is the agentic application / UX layer — the conversational, action-taking front door a business user opens and talks to. Cortex AI is the umbrella suite of underlying services that CoWork (and any custom app) is built on: Cortex Agents (orchestration — plan/tool-call/respond), Cortex Analyst (structured, text-to-SQL over a semantic model), Cortex Search (unstructured, hybrid retrieval), Cortex AISQL / AI_ functions* (primitive LLM ops), plus the newer Cortex Sense (shared context) and the Semantic Models + Horizon governance substrate. The mental model: CoWork/CoCo = the destination app; Cortex AI = the engine room; Cortex Agents = the hinge between them. CoWork is the business-user surface, CoCo (formerly Cortex Code) is the developer surface — two apps over the same services.

How a DSA should frame it in discovery. When a client says "we want Snowflake Intelligence," first translate the name ("that's now called CoWork") and then move the conversation one layer down, because SI/CoWork itself is essentially a toggle — the real engineering is in Cortex. The value phData delivers is not "turn on CoWork"; it is making the layer underneath trustworthy and governed: modeling the semantic view so Cortex Analyst answers structured questions correctly, standing up Cortex Search services over the client's unstructured corpus (with RBAC-aware, metadata-filtered retrieval), and wiring a Cortex Agent that orchestrates both with explicit tool boundaries and eval. A phData-built agent can then be surfaced through CoWork for business users, or through a custom Streamlit/MCP app — same Cortex plumbing, different face. The line to hold: SI/CoWork is what the client's users touch; Cortex Agents + Analyst + Search + AISQL is what phData builds and governs.

Where CAF/Brigade attaches — exactly. CAF (Catalyst Assessment Framework) sits above the entire stack as method: it assesses the estate, classifies use cases, routes by autonomy, and emits a build manifest — it decides which Cortex Agents to build before anyone builds one. It attaches at the front of the funnel (discovery/scoping), not to any single product box. Brigade (the four-station spec→test→code→critic skill-agent) is the factory that produces the build artifacts CAF specifies, and it can execute inside Snowflake's boundary via CoCo (Cortex Code) because CoCo ships a Claude-Code-compatible plugin/skill/subagent model — so the delivery work lands at the Cortex Agents + Analyst + Search layer, in-perimeter, avoiding external Anthropic egress. Net: CAF = the method above the stack; Brigade = the in-CoCo build engine; the deliverable = a governed Cortex Agent; CoWork = the optional business-user face on top.

Open follow-ups

Related

Sources

Vault:

Web: