Databricks has published a practical rubric for choosing which Genie Agents to build first: score five aspects of the workflow, then check whether anyone in the business will champion it. Even a high-scoring candidate should wait if nobody will back its adoption. The useful shift is from asking whether an agent can answer a question to asking whether people repeatedly need that answer to make a decision. An impressive demo is not necessarily an occupied workplace.
Databricks Watch analysis
What happened
In its 2 October guide to choosing first Genie Agents, author Nasir Dakri recommends scoring each candidate workflow from one to five on business impact, recurring demand, data and metadata readiness, scope clarity, and governance and risk fit.
Add the five scores: 20–25 means build now, 14–19 means shape the workflow first, and below 14 means it is not a good fit yet. A business owner or executive sponsor is a separate requirement, not a sixth scoring category. Without an active champion, the advice is to hold off regardless of the total.
The criteria have concrete meanings. Impact means naming the decision, cost or risk the agent would change. Demand means frequent, repeated questions rather than bespoke quarterly analysis. Data readiness means governed tables, certified metrics and useful column descriptions. Scope should cover a tightly bounded domain, while the first deployment should have manageable data sensitivity and regulatory exposure.
The guide also warns against two tempting starts: an “everything agent” covering a whole department, and an executive demo with no recurring use. Other traps include undocumented data, conflicting definitions of the same metric and no named owner to curate and monitor the agent.
Why it matters
This gives teams a way to sort a pilot backlog before spending weeks building it. A repeated inventory question backed by well-defined data may deserve attention before a grand departmental assistant whose remit amounts to “be helpful everywhere”.
It also separates technical performance from adoption. Good answers need sound data and clear definitions; sustained use needs a real workflow and someone prepared to support it. The rubric makes those dependencies visible rather than leaving them for the disappointing post-pilot meeting.
Our read
This is a useful planning tool, not a scientific pass mark. The scoring bands are the author’s recommended framework, and the customer examples illustrate that advice rather than establish a universal predictor of success.
Borrow the questions before borrowing the confidence. Score your candidate workflows, write down the recurring decision each would improve, and identify who will own the agent after launch. If the weak point is metadata, fix the metadata. Another polished demo will not settle what two teams mean by “revenue”.
What to watch
- Whether teams using the rubric report sustained usage and measurable decision-making benefits, not just successful launches.
- How sponsors turn early interest into routine use, and whether agents remain useful when that sponsorship changes.
- Whether data readiness and clear metric definitions prove more informative than the overall score.
Discussion spark: Should a high-value AI-agent pilot be blocked until it has an active business sponsor, or can demonstrated usefulness earn that support after launch?
Sources and evidence
- How to choose your first Genie Agents for maximum impact | Databricks Blog (2 October 2026, 16:15 UTC)
not affiliated with or endorsed by Databricks