The task
Scenario. The ml_features schema is about to feed an agent, and before it does, each table needs a description, an owner, a certification, and a PII classification. You want to check those criteria across the schema and write a readiness score you can filter and track on.
The ask
- Every table in ml_features
- A check of each for description, owner, certification, and PII status
- An AI Readiness score written to custom metadata
Hero prompt
For all tables in the ml_features schema, check if they have descriptions, owners, certifications, and PII classification. Score each one for AI readiness and write the score to the AI Readiness custom metadata field.
Step by step
-
Search for the assets targeted for AI use.
Find all tables in the <ml_features> schema.
-
Check each for description, owner, certification, and PII status.
Check each for a description, owner, certification, and PII classification.
-
Write an AI Readiness score to custom metadata.
Score each for AI readiness and write it to the <AI Readiness> custom metadata field.
What you'll give it
Required context
- The assets targeted for AI use (ml_features)
- The rubric: what "ready" means, and how the criteria weight
- An AI Readiness custom metadata field to write to, and permission to write it
Optional context
- A threshold for "ready" versus "needs work," so the score filters cleanly
- Any criteria to add beyond the four (freshness, a passing quality check)
What Atlan creates
Outputs
- Every ml_features table with its description, owner, certification, and PII status
- An AI Readiness score per table, against your rubric
- The score written to the AI Readiness custom metadata field
Summary. "Is this data fit to feed an agent" becomes a field you can filter, not a judgment call per table. The tables that are not ready are visible, with the reason attached.
Changed: the AI Readiness custom metadata field on each scored table. Left untouched: descriptions, owners, certifications, and the data itself, which are read to compute the score, not edited.
Follow-up prompts
Keep going once the first answer lands.
Show me the gaps
List the tables that scored "not ready" and, for each, the one thing missing.
Fix the easy ones
For tables missing only a description, draft one from the columns so I can approve and raise the score.
Track the trend
Compare this run to last month's scores and show which tables improved and which slipped.
Tips & troubleshooting
Agree on what "ready" means with your team first. The score is only as good as the rubric.
- Match the tool to the work. Scoring is close-reading work; use a more capable model so the rubric is applied consistently across the schema.
- Automate or chain it. Schedule it as the gate before data enters a pipeline: score nightly, and let downstream jobs read the readiness field.
- Ask what it changed. Add "list every table you wrote a score to and the score you wrote." That is the run's record.