Databricks' AI can learn your team's way of doing monthly reports

A new Databricks guide shows how to turn a repeated task into a written 'skill' in Genie One, so the AI does it the same way every time and the whole company can reuse it.

Diagram: a request for last month's trading report goes to Genie One, which picks the matching 'Monthly trading brief' skill from a list and follows its steps, tools, checks and format.
Image: Databricks

The short version

  • Databricks has published a guide to "skills" in Genie One, its AI assistant for company data.
  • A skill is a written recipe for a task your team repeats, such as a monthly trading report, so the AI does it the same way each time.
  • You build one by describing the task in a chat. Then you test it like a new hire's first week.
Screenshot of a chat in which a user asks Genie One to build a monthly trading brief skill, and Genie One replies with numbered questions about metrics, breakdowns and what counts as unusual.
Genie One interviews the user before writing the skill. · Image: Databricks

What Databricks published

On 16 November, Databricks published a detailed guide to skills in Genie One. Genie One is its AI assistant that answers business questions from a company's own data.

Genie One already knows what your business terms mean, such as "net sales". A skill adds how your team likes the work done. A skill holds the steps and checks, plus the format of the result. Ask for "our trading report for last month" and it returns the report in the agreed shape every time.

Genie One uses a skill when you call it by name. It also uses one when it decides the skill fits your request. Each skill is saved as plain-language text that people can read and edit.

How you build one

Databricks suggests a skill whenever you find yourself explaining the same thing to the AI twice. Good signs are a task you repeat often, or one where the format really matters.

Before you start, you answer a few questions. When should it be used? Which data and tools does it need? In what order? What gets checked every time? What should the result look like, and where should a person review it?

Then you describe the task to Genie One in your own words and ask it to interview you. It drafts the skill from your answers.

What the example shows

The guide follows a finance team's monthly trading brief. On the first run, the skill made two mistakes. It reported on September while that month's data was still incomplete. It also flagged product lines too small to matter.

Both were fixed in one message. The team told it to report only on complete months and to ignore anything under $50,000 in sales. Genie One updated the skill and reran the brief. In a fresh chat, it picked August, flagged the two changes big enough to matter, and found emails in the team's inbox that explained each one.

Databricks also recommends saving a "golden" example of a good result inside the skill. That gives the AI a clear target and gives you a way to check it later.

What it means for companies in Indonesia

Most of a company's know-how lives in people's heads and old emails. When a senior analyst leaves, the way they built the board pack often leaves too.

Skills are a practical way to write that know-how down once and reuse it. Once a skill works, you can publish it so other teams can switch it on, with controls over who can edit it. It can also run on a schedule, such as the first working day of each month.

The guide's own advice is the right one: start small. One skill that does one job well is easier to test and fix than one that tries to do everything.

This matters beyond Databricks. Skills are saved in an open format, so other AI assistants that support it can use the same recipe. The work of writing down how your teams operate pays off whichever tool you pick.

What to do this month

  • Ask each team lead for one report they explain to someone every month.
  • Write down its steps and checks before you touch any AI tool.
  • Build it as a skill and test it against a past report you trust.
  • Decide who owns each skill and who may edit it, as you would for a shared spreadsheet.

Sources

  1. 01 · Databricks · 2026-11-16

Keep reading

  1. Analysis

    Korea's GS Caltex now gets new AI models running in hours, not a week

  2. Release

    Indonesia's offices will share records so you stop filling the same forms

  3. Release

    ChatGPT can now answer questions using company data kept in Databricks