Databricks Genie - ramp plan and credentials

Why this is in the vault

Kwik Trip is consolidating on Databricks (see [[2026-08-04-kwik-trip-ai-architecture-engagement-context]]) and asked for a Genie expert; the founder's depth is Snowflake Cortex. This note is the fast path in and the credential plan. Re-read it before any client conversation that names Genie products: the product was renamed twice between April and July 2026, so check the release notes first. Source bias: almost everything here is Databricks-authored (documentation, release notes, blogs, Academy pricing), and the two Cortex-versus-Genie comparison posts (colrows, typedef) are from vendors with a product stake in the semantic-layer space; treat their framing as argument, not measurement.

1. What Genie is now (August 2026)

Official release notes: https://docs.databricks.com/aws/en/ai-bi/release-notes/2026 (every dated item below without its own link is from this page). Abbreviations used: GA = general availability; API = application programming interface; MCP = Model Context Protocol; DBU = Databricks Unit (the billing unit); UC = Unity Catalog; SME = subject-matter expert; UAT = user acceptance testing.

Genie Agents (renamed from "Genie spaces" on 2026-07-09). The curated natural-language-to-SQL object over Unity Catalog tables and metric views. Config surface: general instructions, example SQL, SQL expressions and functions, trusted assets, benchmarks, monitoring. Limit 30 tables/views; docs recommend starting with five or fewer. "Dimensions" renamed "Fields" 2026-05-21. Best practices (updated 2026-07-17): https://docs.databricks.com/aws/en/genie/best-practices : SQL expressions and example SQL first, text instructions "only as a last resort".
Agent mode (was "Research Agent", marketed earlier as "Genie Deep Research"): multi-step plan, query, iterate, report with citations. GA 2026-07-02; APIs in Beta. https://docs.databricks.com/aws/en/genie-agents/agent-mode
Genie Agents API (the Conversation API) GA 2026-04-02; reasoning traces, comments, conversation sharing via API; trusted-asset answers detectable in query.parameters; OAuth user-to-machine (U2M) and machine-to-machine (M2M). https://docs.databricks.com/aws/en/genie-agents/conversation-api
Genie One (was Databricks One, renamed "Genie" 2026-04-27, then "Genie One" 2026-06-09). Account-level chat across Genie Agents, dashboards and apps, with connectors to Google Drive, SharePoint, Salesforce, Jira, Gmail and Microsoft 365. Chat and Documents GA 2026-06-15; Slack and Teams app 2026-06-11; Managed MCP Server Beta 2026-05-28; MCP writes Beta 2026-08-13; scheduled tasks (daily maximum). Data + AI Summit (DAIS) 2026 announcement: https://www.databricks.com/blog/introducing-genie-one-genie-ontology-and-genie-agents
Genie Ontology (Public Preview, on by default since 2026-06-04): auto-extracted context graph from tables, queries, dashboards, pipelines and connected apps, layered over Unity Catalog semantics (metric views, domains). Curation currently free.
Genie Code: agentic coding for dashboards and notebooks (GA 2026-05-28); Power BI and Tableau import GA 2026-08-20; connects to MCP servers.
Metric views: dashboard-scoped local metric views GA 2026-07-30; low-code UI for Unity Catalog metric views.
Scale claim: "1.5M+ Genie Spaces created in 2026" (Databricks blog, 2026-04-26, https://www.databricks.com/blog/next-generation-databricks-genie).

2. Cortex to Genie translation

Cortex concept	Databricks equivalent	The difference that bites
Semantic model YAML / semantic views	Unity Catalog (UC) metric views (MEASURE syntax) plus Genie Agent config (descriptions, synonyms, SQL expressions)	No single authored artifact is "the" semantic layer; context is split across UC metadata, metric views, agent instructions and the auto-inferred Ontology. Ontology is machine-first, the inverse of Snowflake's human-first posture.
Verified queries	Trusted assets: parameterized example SQL plus UC SQL functions and table-valued functions; answers labelled "trusted"	Only parameterized example SQL becomes a trusted asset Genie can adapt; hardcoded filters are just few-shot hints. https://docs.databricks.com/en/genie/trusted-assets.html
Cortex Analyst evals	Benchmarks tab (question plus one ground-truth SQL; judge compares output and shape)	Built into the product; re-run after every config change is the expected loop.
Cortex Search	Vector Search / managed "AI Search" via MCP, plus Genie One document connectors	Genie Agents are structured-only; unstructured goes through Genie One or a custom agent (Genie One chat doc via the release notes above).
Cortex Agents	Agent Bricks / Mosaic AI Agent Framework; a Genie Agent is itself a tool via managed MCP	Orchestration lives outside the Genie config.
Snowflake Intelligence	Genie One	Genie One spans workspaces and external apps; the Genie Agent is one source among several.
Per-message pricing	Genie free for users through 2027-01-31, then 150 DBU per user per month allowance plus SQL warehouse cost	The warehouse runtime is the real bill; service principals (API and automation) are billed today. https://docs.databricks.com/aws/en/genie/monitor-cost

The "machine-first versus human-first" and "orchestration lives outside the config" readings in the table are synthesis from two vendor posts, both with a stake in the space: https://colrows.com/blogs/cortex-analyst-vs-genie/ (updated 2026-08-21) and https://www.typedef.ai/blog/cortex-sense-vs-genie-ontology-which-one-checks-the-answer. The vault's own counter-read on Ontology is [[2026-06-19-data-engineering-central-databricks-summit-2026]]; the Cortex side is [[2026-07-08-cortex-sense-semantic-layer-wedge-caf]] and [[2026-07-07-snowflake-intelligence-vs-cortex-ai-boundary]].

3. Fastest ramp, in order

Read (2 h): the best-practices doc, trusted assets, Agent mode, Conversation API; and two Databricks blogs: "From Data to Dialogue: best-practices guide for high-performing Genie spaces" (2026-02-05, https://www.databricks.com/blog/data-dialogue-best-practices-guide-building-high-performing-genie-spaces) and "How to build production-ready Genie spaces and build trust along the way" (2026-02-06, https://www.databricks.com/blog/how-build-production-ready-genie-spaces-and-build-trust-along-way).
Hands-on in Free Edition (4-6 h): serverless only, Genie Agents and Agent mode available (limits: ~20 questions/min UI, 5/min API; https://docs.databricks.com/aws/en/getting-started/free-edition-limitations). Use the samples.bakehouse dataset (franchise retail: transactions, customers, suppliers, reviews). Build one metric view, an agent on five tables, 10-15 benchmarks, parameterized example SQL; iterate to above 85 percent.
Genie Workbench (3 h): open-source Databricks App from Field Engineering, Create / Score / Fix / Auto-Optimize with MLflow tracking; new agents typically start near 54 percent and climb past 85. https://github.com/databricks-solutions/databricks-genie-workbench
Patterns skill (1 h): the databricks-solutions Genie-space-patterns skill (v2.6, 2026-04-16): metric views first, instructions at most 20 lines, benchmarks with fixed dates, config-drift audits via API. https://github.com/databricks-solutions/vibe-coding-workshop-template/blob/main/data_product_accelerator/skills/semantic-layer/03-genie-space-patterns/SKILL.md
Academy: "Building Reliable Conversational Agents with Genie" (4 h; $750 list for instructor-led; https://www.databricks.com/training/catalog/building-reliable-conversational-agents-with-genie-5217); "Data Analysis with Databricks" (8 h, $1,000 list, certification prep; catalogue entry not captured, price unverified, see below); free "Analytics Fundamentals" badge (~1 h, covers AI/BI and Genie; https://www.databricks.com/resources/learn/training/analytics-fundamentals-accreditation). Partner Academy pricing for phData not verified.
Practitioner reads: Eli Swanson, "Practical Guide to Genie Space Optimization" (Medium); David Huang, "Genie Agent Optimization Loop with Genie Code" (2026-08, https://medium.com/@hiydavid/genie-agent-optimization-loop-with-genie-code-tools-tips-5b8797d9f53d) and his databricks-agent-skills/genie-code/optimize-genie-space.

4. Credentials, ranked by what they signal to a client

Databricks Certified Data Analyst Associate. $200, 45 questions, 90 minutes, two-year validity; exam guide October 2025; domain 7 is "Developing, Sharing, and Maintaining AI/BI Genie spaces" (12 percent) plus metric views. The only certification that names Genie. https://www.databricks.com/learn/certification/data-analyst-associate . Time estimate: a single associate-level exam with a published guide; two to four weeks of evening study is the founder's usual pace for a vendor associate exam (his Anthropic foundations certification took a similar window, see [[2026-06-11-phdata-anthropic-partner-tier-cert-leverage]]).
Analytics Fundamentals and AI Agent Fundamentals free badges, one to two hours each, listed on the certification hub https://www.databricks.com/learn/training/certification . Same-day to earn and post; a badge, not a certification, so it signals currency rather than depth.
Generative AI Engineer Associate. $200, 45 questions per the certification hub above; exam guide dated March 2026; retrieval-augmented generation (RAG), Vector Search, agent evaluation. Genie reportedly not in the guide (unverified against the PDF).
Context Engineer Associate (new, listed as Beta on the certification hub above): context design and governance for agents; most on-theme with Ontology, but too new to have a track record as a client signal.
Partner track: Partner Solutions Architect Essentials badge, then Solutions Architect Champion (requires Lakehouse and GenAI Fundamentals, one Associate plus one Professional certification, three client implementations, panel defence; nomination via phData's partner solutions architect; community reports five-month lags; https://community.databricks.com/t5/certifications/solutions-architect-essentials-badge-champion-candidate/td-p/114069). The program's 2026 status is unclear, and it is a months-scale play, not weeks.

Data Engineer, Machine Learning and Spark certifications: no Genie content, low relevance to this ask.

5. What a client "Genie expert" is asked to do in 2026

Distilled from the two Databricks best-practice blogs in section 3 (Solomon 2026-02-05; Pareek and Lind 2026-02-06), the best-practices doc, the Genie Workbench README and the Genie-space-patterns skill.

Curate a gold layer: pre-joined tables, primary and foreign key constraints, column comments, value dictionaries.
Author Unity Catalog metric views as the trusted-asset spine; publish agents on them, tables as fallback.
Build a 10-20 question SME benchmark set before UAT; report baseline-to-tuned accuracy as the trust artifact.
Tune in order: metadata, then example SQL and SQL expressions, text instructions last; keep instructions short.
Govern: Unity Catalog permissions flow through; Genie One workspace instructions; monitor tab and audit logs; version agent config JSON via API or Asset Bundles and detect drift.
Wire distribution: Genie One in Slack and Teams, dashboard "Ask Genie", iframe embedding (GA 2026-06-04, release notes), MCP exposure to custom agents.
Cost governance: budgets, service-principal billing, warehouse sizing (serverless required for interactive latency).
Adoption: SME champions, a feedback loop, scheduled tasks and alerts for recurring questions.

6. What embarrasses 2024-2025 knowledge

Saying "Genie space" (now Genie Agent), "Databricks One" (now Genie One), "Research Agent" or "Deep Research" (now Agent mode, GA), "Dimensions" (now Fields).
Assuming Genie is free or bundled: pay-as-you-go announced 2026-07-08 (150 DBU per user per month), then a promotion making Genie One and Genie Agents free through 2027-01-31; the GENIE_FREE_USAGE billing SKU (stock-keeping unit) appears from 2026-07-20; service principals are billed (https://docs.databricks.com/aws/en/genie/monitor-cost).
Treating instructions as the primary lever; the docs now say SQL expressions and example SQL first, text last.
Not knowing Genie Ontology, or that Cortex Sense is Snowflake's answer (private preview mid-July 2026 per the typedef post above; see [[2026-07-08-cortex-sense-semantic-layer-wedge-caf]]).
Table limit is 30; scheduled tasks are capped at daily; Agent mode has cross-geography restrictions outside the Americas, the EU, Australia and New Zealand, and Japan (Agent mode doc above).

Not verified

Academy course pricing under the partner program, and the $1,000 list price for Data Analysis with Databricks; whether the Generative AI Engineer March 2026 guide mentions Genie; a standalone Genie Ontology docs page (content taken from release notes and the Genie One chat doc); current status of the Partner Champion program.

RDCO mapping

Feeds the Kwik Trip reference-architecture registry of approved options (Intelligence layer: Copilot Studio and Genie side by side; engagement context in [[2026-08-04-kwik-trip-ai-architecture-engagement-context]], value model in [[2026-08-21-kwik-trip-roi-operating-model]]) and the founder's phData positioning as the cross-platform semantic-layer person (Cortex plus Genie). The Data Analyst Associate is a candidate for the cert-escalator conversation recorded in [[2026-06-11-phdata-anthropic-partner-tier-cert-leverage]].