06-reference/research

snowflake intelligence usage history sql layer

2026-11-08·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
snowflakecoworkaccount-usagebehavior-changerunbooks

The SQL Layer Did Not Follow the Rename: SNOWFLAKE INTELLIGENCE Is Still the Live DDL Keyword, No Bundle Retires It, and the Only Real Break Is Three Renamed Columns

The question

"Is SNOWFLAKE_INTELLIGENCE_USAGE_HISTORY still queryable as an ACCOUNT_USAGE alias, and does any Snowflake behavior-change bundle schedule retiring the SNOWFLAKE INTELLIGENCE DDL keyword or object name?"

Context: this closes the first open follow-up of [[2026-09-15-snowflake-intelligence-vs-cowork-naming-standard]], which established that the product name is retired in docs and billing but left the SQL layer unresolved. It decides whether phData client runbooks and GRANT scripts that name the old identifier need a migration note.

Headline: Three layers, three different answers. The product name is retired (already established, not re-derived here). The view name is gone from current docs in both ACCOUNT_USAGE and ORGANIZATION_USAGE, with no deprecation notice anywhere, and three of its columns were silently renamed. The DDL keyword, object type and privilege names are still SNOWFLAKE INTELLIGENCE in today's docs, verbatim, across create, alter, show, grant, tagging and budget surfaces. No behavior-change bundle schedules any retirement: all eight 2026 monthly bundles contain zero references to this product, keyword or view. The migration note belongs on cost-monitoring queries only. GRANT scripts naming SNOWFLAKE INTELLIGENCE are currently correct and should not be touched.

What we already know (from the vault)

What the web says

All items below verified against docs.snowflake.com on 2026-11-08.

Convergences and contradictions

Synthesis for RDCO

The useful finding is that the three layers moved at three different speeds, and the layer most likely to be in a client runbook is the layer that did not move at all. Snowflake's rename propagated through a documentation substitution variable, which is visible in the raw Markdown as an unresolved %sf-intelligence% token still carrying the old name as its variable key. That mechanism rewrites prose, page titles and even historical release-note headlines in one pass. It cannot touch a SQL identifier, because identifiers are literal text in code blocks and literal grammar in the engine. So the marketing layer flipped completely, the prose around the SQL flipped completely, and the SQL itself flipped only where someone edited a code block by hand. The view name and three of its columns got that hand edit. The DDL keyword, the object type, the three privilege names, the budget domain literal, the SERVICE_TYPE and ENTITY_TYPE values, the tag level enum, and two account-parameter values did not.

For the Principal/PM-track consulting work this matters in a specific and slightly counterintuitive way: the GRANT scripts are fine and should be left alone. A runbook that says GRANT USAGE ON SNOWFLAKE INTELLIGENCE SNOWFLAKE_INTELLIGENCE_OBJECT_DEFAULT TO ROLE ... is quoting today's documentation correctly. Rewriting it to a CoWork spelling would produce a script that fails, and would be the worse error of the two. The defensible migration note is narrow: it goes on cost-monitoring and chargeback queries, where the view name changed and SNOWFLAKE_INTELLIGENCE_ID, SNOWFLAKE_INTELLIGENCE_NAME and SNOWFLAKE_INTELLIGENCE_TAGS became SNOWFLAKE_COWORK_*. Those column renames break a SELECT list even in the generous case where the view name still aliases, which makes them the one unambiguous action item in the whole question.

The behavior-change answer is worth carrying with real confidence because it was tested the right way. Snowflake has a documented channel for SQL-identifier renames and has used it repeatedly, including for a privilege rename and a column rename in output. Nothing about this object sits in any 2026 bundle or in the unbundled list. Combined with the policy's minimum eight weeks of admin-controlled testing and opt-out, and with 2026_08 being the newest bundle, there is no path by which the SNOWFLAKE INTELLIGENCE keyword or object name becomes unusable before the 2027 bundle cycle. If and when it does go, the Code Bundles precedent is the likeliest shape: a published note saying the old statements continue to work and no migration is required.

There is one genuinely odd observation to hold loosely rather than act on. The view rename is the only identifier change in this area that shipped with neither a release note nor a bundle entry, while its direct sibling SNOWFLAKE_COCO_USAGE_HISTORY got a dated GA note. Either the engine still answers to both names, which would make the doc change non-breaking and the missing paperwork correct, or a breaking ACCOUNT_USAGE change shipped outside Snowflake's own stated process. That ambiguity is the reason the view-name half of the recommendation is held at medium confidence while the keyword half is high. It is also a small, concrete credibility artifact: noticing that a vendor's rename skipped its own change-management channel is the kind of observation that earns trust in a discovery conversation, and it fits the agents-in-production domain the founder is building credibility in.

Recommendation: Add a migration note to client cost-monitoring and chargeback queries only, covering the view rename plus the three SNOWFLAKE_INTELLIGENCE_* to SNOWFLAKE_COWORK_* column renames, and explicitly leave DDL and GRANT scripts naming SNOWFLAKE INTELLIGENCE unchanged with a dated line saying why. Confidence: high on leaving DDL and GRANT scripts alone and on no scheduled retirement; medium on the view-name half, which needs one live query to settle.

Why this is in the vault

It closes the first open follow-up of [[2026-09-15-snowflake-intelligence-vs-cowork-naming-standard]] and converts that brief's "don't assume the old view still works" warning into an actionable, dated split: patch usage queries, leave grants alone. It is the citation a phData client runbook or SOW reviewer needs when asked why a 2026 deliverable still writes SNOWFLAKE INTELLIGENCE in SQL while writing "Snowflake CoWork" in prose.

Open follow-ups

Test plan, not a research question: one statement, SELECT 1 FROM SNOWFLAKE.ACCOUNT_USAGE.SNOWFLAKE_INTELLIGENCE_USAGE_HISTORY LIMIT 1, settles whether the old view name still resolves; a successful run should be followed by DESCRIBE to see which column spellings that name returns. Stating this explicitly because it is not researchable: it needs a live Snowflake account with ACCOUNT_USAGE access, which we may not have, so it belongs to whoever next has a client or trial account in hand rather than to the research queue.

Related

Sources

Vault:

Web, primary (Snowflake documentation, all retrieved 2026-11-08):

Web, archived primary (dated capture of Snowflake documentation):

Web, corroboration only (not used as the finding):