Task 1.3: Define and order tools that gather knowledge or take actions for multi-stage reasoning; Determine how and when to use Agent Bricks (Knowledge Assistant, Multiagent Supervisor, Information Extraction) to solve problems

• AOne Unity Catalog function that runs any SQL text the model writes, used for both kinds of request

  Incorrect

  Unity Catalog function tools fit a query that is known ahead of time and takes parameters from the agent, not SQL that the model writes.

• BThe Genie One MCP for both kinds of request, including the employee leave-balance lookup

  Incorrect

  Genie One fits open-ended business questions, but the balance must come from the approved query, which belongs in a parameterized function.

• CA Unity Catalog SQL function on the employee ID for balances, and the Genie One MCP for the rest

  Correct

  Databricks maps a predefined query with parameters to Unity Catalog function tools, and natural-language business questions across the workspace to the Genie One MCP.

• DThe Databricks SQL MCP for both kinds of request, so that the model writes and runs every query

  Incorrect

  That MCP is for writing and running SQL, so it would bypass the approved balance query, and system.ai.dbsql also needs the Unity Gateway beta.

Choose a structured-data tool by who writes the query. A Unity Catalog SQL function runs a known query with parameters that the agent supplies; the Genie One MCP interprets business questions across the workspace and queries the data with the caller's permissions.

• AKnowledge Assistant pointed at the invoice volume

  Incorrect

  Knowledge Assistant is a legacy document chatbot; it answers questions rather than filling a schema, and it is not recommended for new agents.

• BInformation Extraction, then ai_extract in a scheduled pipeline

  Correct

  Information Extraction defines a schema, scores each field against a ground-truth table, and hands off to ai_extract in SQL or a Lakeflow pipeline for scheduled runs.

• CSupervisor Agent coordinating one extractor per field

  Incorrect

  Supervisor Agent is a legacy multi-agent coordinator and adds orchestration that a single schema extraction does not need.

• DA Genie Agent over the Unity Catalog table that lists the invoice files and paths

  Incorrect

  A Genie Agent answers natural-language questions with SQL over curated tables; it does not read field values out of PDFs.

Information Extraction turns unstructured documents into structured fields with a defined schema, evaluates extraction quality against labeled data, and runs at scale through ai_extract. Knowledge Assistant and Supervisor Agent are legacy offerings.

• AIt is in Public Preview and open to new projects through the Previews page

  Incorrect

  It is not in preview; it has been retired.

• BIt reached end of life and is no longer available

  Correct

  The Supervisor API reached end of life on September 30, 2026; Databricks recommends writing your own agent loop.

• CIt is generally available and recommended

  Incorrect

  It is no longer available at all.

• DIt moved into Genie One unchanged

  Incorrect

  There is no such move; the API simply ended.

The Supervisor API was deprecated and then reached end of life on September 30, 2026. Databricks recommends writing your own agent loop and deploying it.

• ADocument Parsing

  Correct

  Document Parsing shows the source document next to the parsed result and can hand off the ai_parse_document query.

• BClassification

  Incorrect

  Classification assigns documents to labels; it does not display parsed layouts.

• CInformation Extraction

  Incorrect

  Information Extraction fills a schema of fields, not a visual parse of the layout.

• DGenie Agents

  Incorrect

  Genie Agents answer questions over data with SQL.

Document Parsing extracts and visualizes structured content from PDFs, images, Word and PowerPoint files and offers the ai_parse_document query to run at scale.

• AKeep the single label and turn on confidence scores so both cases are reported

  Incorrect

  One label is not enough to run classification, and confidence scores do not add labels.

• BSwitch to Information Extraction

  Incorrect

  Extraction fills schema fields; it is not label assignment.

• CAdd at least one more label and enable Multiple labels

  Correct

  Classification needs at least two labels to run, and the Multiple labels option lets it return more than one label.

• DEnable rationales to add labels

  Incorrect

  Rationales explain predictions; they do not add labels.

In Classification you add labels manually or import them from a table, need at least two labels, and can enable Multiple labels, confidence scores and rationales.

• ARegister the Genie Agent as an external MCP server in Unity Gateway

  Incorrect

  Registering Genie as your own MCP is not supported; use a Databricks-provided MCP or AI Bridge.

• BExport its example SQL into the prompt

  Incorrect

  Static SQL in a prompt loses Genie's interpretation of new questions.

• CGive the agent a raw JDBC connection

  Incorrect

  Raw SQL access skips the curated context that makes the Genie Agent reliable.

• DWrap AI Bridge's GenieAgent helper as a tool

  Correct

  AI Bridge exposes a specific Genie Agent as a Python tool and handles requests, polling and results through the Genie API.

Databricks picks structured-data tools by what the agent needs and who writes the query; for one specific Genie Agent in a Python agent, AI Bridge wraps the Genie API.