Skip to main content

Decision brief

Looker Review: Governed BI and Gemini Agents on One Semantic Layer

Choose Looker when dashboards and AI answers must share one LookML model and a data team will keep it current. This review covers licensing, data tokens, and data handling before Gemini reaches business users.

Reading time
3 min read
Sources
19 cited
In this piece
Looker−8 Mixed

GILD verdict

Choose Looker when reports and AI answers must share one semantic layer, and a data team will keep LookML current.

Google's internal testing found that layer cut data errors in generative AI queries by up to two thirds. Assign LookML maintenance before opening Conversational Analytics, and measure data token use while Google's promotion lasts.

Looker

AI Data Analysis by Google

−8 Mixed
Product page

Fit and limits

Who it fits

  • Your data is modeled in BigQuery, Snowflake, Databricks, Redshift, or another supported SQL database, and the data team writes SQL.
  • Departments argue about whose number is right, and you want each metric, join, and access rule defined once in LookML and reviewed in Git.
  • Business users need answers in Looker, Google Sheets, Excel, Power BI, or Gemini Enterprise, all reading the same model through connectors and published agents.
  • You want AI answers that respect existing permissions: Conversational Analytics composes queries through LookML, so access grants and user attributes apply.

Where it falls short

  • Every answer depends on LookML: labels, descriptions, hidden fields, and custom fields promoted into the model decide whether agents pick the right field.
  • Pricing runs through sales on annual terms with no published list price. Practitioners call the license expensive, including next to Power BI, and report dashboards that drive warehouse compute.
  • Chart and formatting options are narrower than Tableau or Power BI in many public accounts, and some teams moved business-user dashboards to Sigma or Domo.
  • Several AI features are pre-GA, including dashboard agents, agentic workflows, and the managed MCP server, and Pre-GA features are not recommended for regulated workloads.

Implementation reality

  1. Curate one or two Explores

    Pick Explores on clean data with clear business value. Give every field a business label and description, hide keys and technical fields, add synonyms, and move commonly used custom fields and table calculations into LookML so agents can use them. Google's Looker extension for VS Code lets an AI agent such as Claude Code generate, edit, and validate LookML for a user who holds the develop permission on that model.

  2. Build data agents and test them

    Create specialized agents for one domain each, connected to no more than five Explores, with instructions, default filters, and verified queries for the questions that matter. Test them with people who know the data before anyone else sees them.

  3. Grant access by model

    Give the gemini_in_looker and access_data permissions only on curated models through a custom role. Review the Gemini Default Users group, which grants the Gemini role automatically on Looker (original) instances with open system configuration.

  4. Baseline usage before quotas

    Watch the Conversational Analytics dashboard in System Activity for engagement and estimated token use while the promotion lasts. Its Token usage tab estimates input and output tokens by day and by agent. Input covers each question, the session history, and the agent's metadata and instructions. Budget warehouse compute for the extra queries agents and dashboards send.

What to measure in your pilot

  • A pilot admin reports about 250,000 input tokens per question, so Standard's 60M pool covers about 240 questions; divide your own System Activity input tokens by questions asked.
  • GILD found no cross-domain accuracy figure in Google's docs; score one verified-query set per agent, then rerun it on a combined agent before merging domains.
  • Track how many LookML changes AI agents draft per sprint and how many fail validation or review; keep a developer approving every merge until fewer than 1 in 10 fail.

Sources

Public source
1
Official documentation
18
  1. Google Developer forums: Conversational analytics / gemini token usage data (Sep 14, 2026)Public sourcediscuss.google.dev
  2. LookerOfficial documentationcloud.google.com
  3. Looker pricingOfficial documentationcloud.google.com
  4. Looker docs: Looker (Google Cloud core) features and entitlementsOfficial documentationdocs.cloud.google.com
  5. Looker docs: Choosing a hosting option for a Looker (original) instanceOfficial documentationdocs.cloud.google.com
  6. Looker docs: Gemini in Looker overviewOfficial documentationdocs.cloud.google.com
  7. Looker docs: Administer Gemini on your Looker (Google Cloud core) instanceOfficial documentationdocs.cloud.google.com
  8. Looker docs: Admin settings, Gemini in LookerOfficial documentationdocs.cloud.google.com
  9. Looker docs: Conversational Analytics in Looker overviewOfficial documentationdocs.cloud.google.com
  10. Looker docs: Set up Conversational Analytics in LookerOfficial documentationdocs.cloud.google.com
  11. Looker docs: Best practices for configuring Conversational AnalyticsOfficial documentationdocs.cloud.google.com
  12. Looker docs: Create and manage Explore data agentsOfficial documentationdocs.cloud.google.com
  13. Looker docs: Looker-managed MCP serverOfficial documentationdocs.cloud.google.com
  14. Looker docs: Admin settings, BI ConnectorsOfficial documentationdocs.cloud.google.com
  15. Google Cloud docs: How Gemini products in Google Cloud use your dataOfficial documentationdocs.cloud.google.com
  16. Looker release notesOfficial documentationdocs.cloud.google.com
  17. Google Cloud blog: How Looker's semantic layer enhances gen AI trustworthiness (May 7, 2025)Official documentationcloud.google.com
  18. Looker docs: Admin settings, System Activity dashboardsOfficial documentationdocs.cloud.google.com
  19. Looker docs: AI-assisted development (vibe coding) with LookerOfficial documentationdocs.cloud.google.com

Keep reading

  1. Buyer guideMoving Dashboards From Looker or Tableau to HexHow to audit, rebuild, and retire Looker or Tableau dashboards in Hex, what the migration agent does, and what the data team still checks.
  2. Buyer guideLooker Conversational Analytics Rollout PlaybookHow to curate Explores, build and test data agents with verified queries, and widen access to Gemini in Looker one domain at a time.
  3. Buyer guideLooker Licensing & Data Token Budget ChecklistWhich edition and user licenses you need, how far the monthly data token pool goes, and how to measure Conversational Analytics usage before overage billing starts.
Read the research