Should Atlan auto-publish AI-generated asset descriptions or require steward approval first?

Introduction

Atlan AI and Atlan's context agents can generate descriptions for tables, columns, dashboards and READMEs at scale. For a customer with 200,000 assets and a governance team of five, this is the difference between a catalogue that stays mostly empty and one that is useful. The question is whether AI-generated descriptions should be auto-published, visible to everyone immediately, or require a steward's approval before they appear.

Auto-publishing maximises coverage and speed. Requiring approval maximises accuracy and trust. In a governance product, trust is the product, so a wrong description that misleads an analyst or an AI agent is costly. But an empty catalogue is also costly, and stewards cannot review everything.

Step 1

Clarifying Questions (3 minutes)

• Context: What share of assets are documented today at a typical customer?

• Hypothetical answer: About 20–30% of assets; top-used assets around 50%.

• Business context: What matters more for renewals: coverage or accuracy?

• Hypothetical answer: Both; governance buyers emphasise trust, but low coverage drives low adoption.

• User impact: How accurate are AI drafts today?

• Hypothetical answer: In sampled reviews, about 80% are acceptable without edits, 15% need edits, 5% are wrong.

• Technical: Can we label AI-generated content and track its status?

• Hypothetical answer: Yes; content can be marked AI-generated, with a separate verified state.

• Resource: How much steward time is available?

• Hypothetical answer: A few hours per steward per week.