Which data analysis and BI jobs can an AI agent do today?

The AI agents for data analysis cited here document specific tasks, including writing SQL, creating visualizations and flagging pipeline issues: they turn a plain-language question into SQL, run it, and return a table or chart, and one watches pipelines and proposes fixes when a job breaks. The catch is that their output can be wrong. Microsoft warns that Copilot can return inaccurate results when its semantic model is not prepared, and OpenAI advises reviewing generated code, outputs and assumptions. So the job you hand over is "query, chart and flag", and the job you keep is "define the metrics and check the number against one you already trust".

This page covers what data agents do according to their documentation, which data jobs you can hire one for, which role changes, and where to start. It describes the products and benchmark cited, as of November 2026.

What AI agents do in data analysis today

The products read for this page fall into three groups.

Answering questions from a warehouse or BI model.

• Snowflake Cortex Analyst. Snowflake's documentation describes it as a managed feature that lets business users "ask questions in natural language and receive direct answers without writing SQL." It relies on a semantic view that defines tables, metrics and join paths, and it runs the generated SQL under Snowflake's role-based access controls. Snowflake lists its limits plainly: it "is limited to answering questions that can be resolved with SQL" and does not answer questions like "What trends do you observe?" Snowflake now recommends moving to Cortex Agents.
• Databricks Genie Agents. According to Databricks, a Genie Agent is a chat interface where users "ask questions of their data and get back SQL queries, results tables, and visualizations." Data analysts curate each one with datasets, example SQL queries and instructions in the company's own terms. Databricks states that Genie One and Genie Agents usage by users is free through January 31, 2027; service principals are billed.
• Copilot in Power BI. Microsoft's documentation says Copilot answers questions about reports and semantic models, summarizes reports, helps authors create report pages and writes DAX queries. Microsoft also warns that without preparing the semantic model, "Copilot can misinterpret the data and return generic or inaccurate results." Some Copilot experiences are generally available and others are in preview.

Analyzing a file you upload. According to OpenAI's help center, ChatGPT can analyze uploaded spreadsheets, CSV and other files, create tables and charts, and run Python in a notebook environment. The same page tells users to "review the generated code, outputs, and assumptions before relying on the result."

Watching pipelines. Databricks documents Genie ZeroOps, in Beta, as a background agent that "monitors your jobs, pipelines, and tables, detects failures and data quality issues, identifies the root cause, and proposes a fix." Databricks says it "doesn't change production data or code unless you approve the change," and that it does not detect performance or cost regressions.

The common thread across the Snowflake, Databricks and Microsoft documents: each product works from something a person set up first (a semantic view, curated datasets and example queries, a prepared model), and Microsoft warns that without that preparation output can be inaccurate. That setup and a check of the output are your part of the job.

How good is the SQL? Spider 2.0 is a research benchmark of 632 text-to-SQL workflow problems taken from enterprise database use cases, with databases that often have over 1,000 columns. The benchmark's site reports that GPT-4o's success rate on Spider 2.0 tasks was 10.1%. On the live leaderboard read for this page, the highest listed score on the 547-task Snowflake set is 96.70, while the highest on the 68-task dbt set, a code agent task that requires understanding project code, is 65.6. The leaderboard labels these figures "Score" without defining the metric on that page, and the dbt-set score is the lower of the two. A benchmark score is not a measure of any product on your data.

Jobs you can hire an agent for

Each job has its own page with a task breakdown, a brief template, risks and how to run a trial.

• JobSupervised
  AI agents for business intelligence and reporting

  An AI agent can answer plain-language metric questions, write the SQL and refresh recurring reports from your governed data. Your data team still defines the metrics, checks numbers that leave the company and owns the conclusions.
• JobSupervised
  AI agents for data engineering: what they do today

  AI agents can draft pipeline code, add data quality checks, write documentation and diagnose failed runs. A data engineer still reviews every change before it reaches production and owns business logic and access.
• JobAssist only
  AI agent for spreadsheet analysis: what it can do

  An AI agent inside Excel can summarize a dataset, add formula columns and PivotTables, flag outliers and trace formula errors. Both vendors cited here tell you to verify the output, so a person checks the numbers before anyone relies on them.

Ready

0of 3

Supervised

2of 3

• AI agents for business intelligence and reporting
• AI agents for data engineering: what they do today

Assist only

1of 3

• AI agent for spreadsheet analysis: what it can do

Not yet

0of 3

They differ in what the agent touches and how you check it:

Job	What the agent does	What a person keeps
Reporting and BI	Answers questions from a governed model, drafts charts and report pages	Metric definitions, which numbers go to leadership
Data engineering	Flags broken jobs and stale tables, proposes code fixes	Approving every change to production code and data
Spreadsheet analysis	Cleans, summarizes and charts an uploaded file	Checking the method and the totals before anyone uses them

For analysis of financial statements and forecasts, see the financial analyst page.

Roles they affect

This function covers one role, the data analyst. Its page goes task by task through what an agent can take and what stays with the person.

• Role
  Can AI replace data analysts? A task-by-task answer

  Documented AI tools clean data, answer questions in SQL or Python, and build charts, with a person reviewing the output. Framing the business question, building predictive models, defining metrics and presenting recommendations have no cited agent evidence and stay human.

BLS has no occupation called "data analyst". Its closest group is data scientists (SOC 15-2051), and the BLS page for that group links O*NET's Business Intelligence Analysts profile under the same code. BLS puts the median annual wage for data scientists at $120,230 (May 2025) and projects employment to grow 35% from 2025 to 2035. The BLS duty list includes collecting and analyzing data and using visualization software to present findings, which overlap with what the products above document. It also includes making business recommendations to stakeholders, which none of the documentation read for this page describes an agent doing.

Where to start

Start with a question you already know the answer to. A wrong query still returns a number, so the first test is whether the agent can reproduce a number your team has already reconciled.

For example (assume, for this example, a 40-person company with sales data in a cloud warehouse): you give an agent read-only access to the sales schema and a definition of "net revenue" that excludes refunds and tax. You ask it for net revenue by month for last quarter. Finance closed that quarter at a known total. A person compares the agent's three monthly figures with the closed books. If they match, you open the SQL and check the joins and filters. If they don't, you find the cause in the SQL (a missing filter, a different definition, a wrong join) and fix the model or the brief before the next question.

Before you hand over any data job, we recommend:

1. Give read-only access first. Two of the products above document that they work within existing permissions: Snowflake runs generated SQL under its role-based access controls, and ZeroOps monitors only assets the configuring user can access. Create a role that can read the tables in scope and nothing else. Grant write access to production data or code only for a named job, with approval on each change (our recommendation).
2. Write the metric definitions down. Revenue, active customer, churn: put each one in the semantic layer or the brief. Snowflake, Databricks and Microsoft all tie answer quality to this preparation.
3. Keep one reconciliation number per job. A figure from closed books, a published report or a manual count that the agent's output must match.
4. Ask for the query or code every time. Cortex Analyst and Genie Agents return SQL, and Copilot in Power BI can write DAX queries. An answer you cannot trace is one you cannot check.
5. Check where the data goes. Snowflake says Cortex Analyst uses Snowflake-hosted models by default and does not train on customer data. Databricks says ZeroOps notifications sent to Slack or PagerDuty sit outside its retention controls and that ZeroOps is not HIPAA compliant. Read the equivalent page for whatever you use.
6. Keep decisions with a person. The agent produces the number; someone who owns the metric decides what it means.

If you want help scoping a data job, post it below. The guides on hiring an AI agent and writing an agent job description cover scope, trials and acceptance criteria.

Sources

1. 1
   Cortex Analyst

   Snowflake DocumentationPrimaryAccessed Nov 14, 2026

2. 2
   Genie Agents

   Databricks DocumentationPrimaryAccessed Nov 14, 2026

3. 3
   Genie ZeroOps

   Databricks DocumentationPrimaryAccessed Nov 14, 2026

4. 4
   Genie

   Databricks DocumentationPrimaryAccessed Nov 14, 2026

5. 5
   Copilot for Power BI overview

   Microsoft LearnPrimaryAccessed Nov 14, 2026

6. 6
   Data analysis with ChatGPT

   OpenAI Help CenterPrimaryAccessed Nov 14, 2026

7. 7
   Spider 2.0: Evaluating language models on real-world enterprise text-to-SQL workflows (benchmark and leaderboard)

   Spider 2.0 team (XLang Lab, University of Hong Kong, and collaborators)PrimaryAccessed Nov 14, 2026

8. 8
   Data Scientists: Occupational Outlook Handbook

   U.S. Bureau of Labor StatisticsPrimaryAccessed Nov 14, 2026