🎉 Recap: Our First Genie Community Office Hours (8 July 2026) - If you attended please complete our survey now: https://surveys.training.databricks.com/jfe/form/SV_3ekm8gNElFRUf30
Thank you to the ~81 of you who joined our very first Genie Community Office Hours! This community exists to connect practitioners and customers of every skill level - to learn together, share best practices, and build champions with Genie. Amazing energy in the chat - so much so that we couldn't get to every question live, so we've captured a summary below.
A huge thanks to the all-star Databricks team for hosting and supporting with all of the Q&A: Christophe Chieu, Elizabeth Abraham, Kevin Risch, Lingeshwaran Kanniappan, Sivakumar Ganesan, and Youssef Mrini.
📚 Training & enablement — AI/BI & Genie Summer Camp - coming week of July 20 - more details to come
- AI/BI for Data Analysts — 4-hour virtual ILT
- Genie Producer — 4-hour session on building conversational agents (SQL proficiency recommended)
- Genie Consumer — 3-hour workshop for non-technical business users
💡 Organizations with 100+ interested participants can request a free, dedicated training session — reach out if that's you. Registration links and the full July schedule will be posted here on the forum.
🧠 Product highlights
- Ontology — a live, self-improving context layer that closes the "business context gap." Using a PageRank-inspired algorithm to rank business rules and definitions, it reportedly lifted AI accuracy from ~50% to 84.5% while reducing token costs. Connects to sources like email, Google, and SharePoint.
- Genie One — a unified environment to ask natural-language questions, schedule tasks, and work with data on mobile. Conversations can be turned into agents shared company-wide via Slack or Teams.
- Gen Code & Gen Builder — natural-language code/pipeline generation and app-building (even from screenshots), with native Unity Catalog integration for table discovery and security.
- Zero Ops — automated debugging: clones pipelines to find/fix errors (pushable via CI/CD) and scans ML models on metrics like precision/recall to flag issues such as overfitting.
- Terminology update: "Genie Spaces" are now "Genie Agents," reflecting new capabilities like task scheduling and automation.
❓ Your questions from the chat & survey
On metadata & accuracy
- Can Genie scan the underlying data to auto-generate accurate metadata instead of relying on existing metadata? (Mani) — Genie uses schema info, sample values, relationships, and Genie Code suggestions to generate context automatically, but it still depends on curated metadata and human review. Better auto-generated metadata improves results, but it doesn't fully replace a well-curated semantic layer.
- Isn't the point of Ontology to remove these "manual" steps? (Gary) — Ontology automates building the knowledge graph and context mapping, but you still input the domain-specific business rules that are unique to your organization.
On managing instructions & business rules
- We're hitting a "Long Text" warning putting non-SQL business rules into the instruction set — does this hurt token costs, and are there alternatives? (Gary) — Keep instructions concise and keyword-driven; split Genies if exceptions get too broad. Use compaction to summarize tokens, and expose complex logic as Python functions via Unity Catalog rather than long free-text.
- Our own benchmark query keeps getting evaluated as "bad," which lowers accuracy — how do we deal with that? — Great one we want to dig into further; we'll follow up on the forum with guidance (see open items below).
On governance, permissions & scaling
- Consumer-permission vs. creator-permission modes can mean two users see different data — can we pin a Genie to a fixed set of assets so everyone consumes from the same source? — Genie One may route queries to different Agents based on a user's permissions. Best practice: one Genie per subject area for a consistent, trusted source across all business users.
- How do we layer an agentic/Genie experience over an existing Databricks App running predefined SQL, and get governed answers without hand-coding every query? (Akash) — Use Unity Catalog Metrics for authoritative, governed responses, then run the loop: deploy to business users → monitor responses → refine instructions/metrics.
- As we scale to hundreds of Spaces and thousands of users, which DE/governance decisions most affect accuracy, performance, and adoption? — Limit tables per Genie (aim for ~5–6), use serverless SQL warehouses, and distribute workloads across workspaces.
On performance
- For very large Delta tables, does Genie automatically use Photon, Liquid Clustering, materialized views, or caching? — Performance is handled by the selected SQL warehouse (Photon, liquid clustering, caching). Adjust warehouse size / increase compute for high concurrency or slow responses.
On usage tracking & budgets
- Genie usage isn't tagged in system.billing.usage — how do I build a monitoring dashboard? (David) — You can track non-billing usage via the Audit access tables (these include Genie One chats; Genie Code has a separate table). Isolate Genie Spaces onto separate compute to loosely link access usage to billing. For spend, the built-in Budget UI tracks usage by user/group and lets you set thresholds — no custom dashboard needed. (Thanks to the community member who shared the Audit-tables tip!)
- We set a budget threshold and see values appearing — does the dashboard show only pay-as-you-go, and does data appearing mean the free monthly allowance is used up? (Alciona) — Correct: the Budget dashboard only shows consumption above your free monthly allowance, so records appearing means you've passed the free tier.
On pricing & free usage
- Is Genie pay-per-DBU / can we use it without paying? Effective 6 July 2026, each user gets 150 DBUs of free LLM usage per month — is there a cap on how many users can use the free tier? (Indra) — There's a free monthly allowance per user; the Budget UI shows consumption above it. We'll confirm the specifics on per-user limits and post the details on the forum.
On migration
- Can we migrate Tableau dashboards (with actions and parameters) to AI/BI? (Sugandha / Indra) — Yes — Genie skills can facilitate the migration, and AI/BI dashboards support parameters. See the import guide in Resources below.
On automation & DABs deployment
- Can Genie detect data accuracy in a table, and is automation possible in a Genie Space? (Artem) — Genie Agents now support task scheduling and automation.
- When deploying Genie Spaces via DABs: JSON config doesn't support variable interpolation, and Spaces aren't bind resources (each deploy replaces the Space and loses history) — any roadmap for these? — This is a great product/roadmap question we're taking back to the team; we'll follow up on the forum.
📎 Resources shared in the session
- 📝 Introducing Genie One, Genie Agents, and Genie Ontology (blog) (https://www.databricks.com/blog/introducing-genie-one-genie-ontology-and-genie-agents)
- 📝 Introducing Genie ZeroOps (blog) (https://www.databricks.com/blog/introducing-genie-zeroops)
- 📘 Genie best practices (https://docs.databricks.com/aws/en/genie/best-practices)
- 📘 Genie Code skills (https://docs.databricks.com/aws/en/genie-code/skills)
- 📘 Audit access tables (usage tracking) (https://docs.databricks.com/aws/en/ai-bi/admin/audit)
- 📘 Unity Catalog metric views (https://docs.databricks.com/aws/en/business-semantics/metric-views/)
- 📘 Migrate Tableau reports to AI/BI dashboards (https://learn.microsoft.com/en-gb/azure/databricks/dashboards/manage/import-bi)
- 🎓 Training: Building Reliable Conversational Agents with Genie (https://www.databricks.com/training/catalog/building-reliable-conversational-agents-with-genie-5222)
🔎 Open questions we're following up on
- The "benchmark query evaluated as bad" accuracy issue
- Per-user limits on the free LLM/DBU allowance
- DABs roadmap: JSON variable interpolation & bind-resource support for Genie Spaces (to preserve history across deployments)
We'll post answers to these here as we get them.
📌 What's next
- 📅 Next Office Hours in a couple of weeks (scheduled to be West Coast-friendly) — we're exploring a weekly cadence based on your demand.
- 🎤 Our first Genie Community event is being planned in partnership with Advancing Analytics, featuring speakers from customer organizations.
- 📝 Please complete the post-session survey (https://surveys.training.databricks.com/jfe/form/SV_3ekm8gNElFRUf30) - it directly shapes what we build next.
The forum is home base - please keep the questions and conversations going here so the whole community benefits. Thanks again, and see you at the next one! 🚀