Both humans and agents depend on trustworthy context to work effectively. Yet, as organizations increasingly adopt data agents—conversational analytics copilots, data engineering assistants, data science companions—many teams are discovering the same uncomfortable truth: agents can generate SQL that looks correct but isn’t aligned with business reality.
This is the story of how Golden Queries in Dataplex Universal Catalog fix that problem.
Meet Asha: An Analyst Navigating the Early Age of Data Agents
Asha is a senior data analyst supporting the product and marketing teams at a fast-growing subscription company. She’s comfortable with SQL but often turns to conversational Data Agents to speed up routine tasks:
-
“Show me churn rate over the last 3 months.”
-
“What percent of users engaged with the new feature?”
-
“Give me a breakdown of active users by geography.”
Most of the time, the agent responds quickly with runnable SQL. But increasingly, Asha notices a pattern: the queries are technically valid yet semantically off.
-
Wrong fact tables
-
Missing business rules
-
Incorrect filters
-
Join paths that don’t reflect team norms
Nothing catastrophic—just wrong in subtle, quiet ways. And in analytics, quiet errors are the dangerous ones.
Why does this keep happening?
The Root Problem: Agents see Schema, Not Intent
LLMs and Data Agents work by interpreting metadata: table names, column names, schema structures, descriptions, and examples.
But in most organizations:
-
Table descriptions are sparse.
-
Column comments are outdated.
-
Business logic lives in documents, slack threads, and memory.
-
“How we actually compute things” is tribal knowledge known by a few.
So when Asha asks:
“How many active users did we have last month?”
The agent confidently generates:
SELECT COUNT(*)
FROM users
WHERE last_active_date >= ‘2024-10-01’
This looks fine—but it’s wrong.
Because inside Asha’s company, “Monthly Active Users” isn’t defined by a single column.
It’s based on:
-
login activity
-
OR payment activity
-
OR any session longer than 5 minutes
-
sourced from a fact_user_engagement table
-
joined with dim_users for user attributes
The agent didn’t know any of this. Why? Because the metadata didn’t tell it.
This is where Golden Queries change the game.
Enter Golden Queries
Golden Queries are user-verified, high-quality SQL queries published into Dataplex Universal Catalog. This is currently in public preview. They capture:
-
canonical business definitions
-
preferred join paths
-
required filters
-
compliance-safe patterns
-
domain expertise
And because they live in the catalog, both humans and agents can discover and reuse them.
Think of them as practical, executable documentation—metadata with teeth.
How Golden Queries help Asha — A Before/After
Let’s revisit the “Monthly Active Users” example.
Before Golden Queries
The agent guesses.
It chooses the users table.
It infers activity from a single timestamp.
It misses the canonical business logic entirely.
After Golden Queries
The team creates a Golden Query for MAU:
SELECT COUNT(DISTINCT user_id) AS monthly_active_users
FROM fact_user_engagement
WHERE (
login_event = TRUE
OR payment_event = TRUE
OR session_duration_minutes > 5
)
AND event_date BETWEEN DATE_SUB(CURRENT_DATE(), INTERVAL 30 DAY) AND CURRENT_DATE()
They publish it to Dataplex Catalog.
Now, when Asha asks the same question:
“How many active users did we have last month?”
The Data Agent retrieves the Golden Query, understands:
-
which table the business uses
-
what “active” means
-
how time windows work
-
how to safely extend it
The agent then adapts the Golden Query into the answer Asha needs—even if the question changes, like:
-
“Active users by geography”
-
“Active users split by acquisition channel”
-
“Weekly active instead of monthly”
Because the business logic foundation is there.
This is the difference between an agent that guesses and an agent that understands.
Golden Queries Shine Beyond Metrics
Golden Queries aren’t just for KPI definitions. They bring clarity to several high-value workflows:
1. Complex Joins
Agents often generate incorrect or inefficient join paths.
Golden Queries show the canonical join pattern:
FROM fact_sales
JOIN dim_users USING (user_id)
JOIN dim_products USING (product_id)
Agents use this as the default instead of inventing joins.
2. Feature Engineering
Data Science Agents rely on examples for:
-
sessionization logic
-
rolling windows
-
derived features
Golden Queries provide these templates.
3. Compliance & Governance
Teams can publish:
-
PII-safe query patterns
-
approved export queries
-
GDPR-compliant filters
Agents learn the safe patterns automatically.
4. Exploratory Analytics
New analysts often ask:
“What’s the best way to start exploring this dataset?”
Golden Queries give examples that reveal structure, intent, and common usage.
How Golden Queries are Created and Published
-
Generated automatically through Data Insights scans
-
Published into Dataplex Universal Catalog as a ‘Queries’ aspect.
-
Edited via UI or API
-
Verified by analysts, engineers, or domain experts
-
Used by both humans and agents as trusted references
Asha can even collaborate with a conversational agent:
“I’ve generated a query for monthly active users—should I publish it as a Golden Query?”
This closes the loop between conversational exploration and shared organizational knowledge.
How Agents use Golden Queries under the Hood
When a user asks a question, agents:
-
Retrieve relevant tables from the catalog
-
Retrieve Golden Queries associated with them
-
Learn the patterns embedded within
-
Generate fully contextualized SQL based on those patterns
In other words, Golden Queries act as real-time grounding signals for agentic reasoning.
The result: Agents become not just capable, but trustworthy.
The Organizational Impact
With Golden Queries in place:
-
Metric definitions align across teams
-
Dashboards become consistent
-
Analysts onboard faster
-
Agents hallucinate less
-
Self-service becomes safer
-
Data literacy increases organically
-
Business context becomes institutional, not tribal
Asha feels this immediately. Her conversations with agents become more accurate, more predictable, more confident.
Instead of trust but verify, she moves toward trust and extend.
Building AI that Understands your Business
Golden Queries turn business logic into metadata. They allow agents to reason with intent, not inference. And they transform stray SQL patterns into reusable organizational knowledge.
As organizations adopt more agents—1P, 3P, and domain-specific copilots—the Universal Catalog becomes the foundation they all learn from.
With Golden Queries, that foundation becomes richer, clearer, and more aligned with how the business actually works.
Better metadata → better agents → better decisions.
Golden Queries make that loop real.