A few years ago I wrote about self-service BI: the promise that any user could explore data without depending on IT. That promise was only half kept, because it still required knowing how to build reports. In 2026, agentic analytics picks up that idea and takes it to the next level: instead of building a report, you ask an agent.
What is a data agent?
A data agent is an AI-based assistant that understands natural language questions, queries your governed data sources and returns an answer, often with a visualisation, a summary and even recommendations. Unlike a generic chatbot, it works on your data and respects your security rules and business definitions.
The key difference with traditional BI is the starting point. Before: open the tool, choose dimensions and metrics, build the chart. Now: type “compare this quarter’s sales with the previous one by region” and let the agent do the rest.
How each platform tackles it
The three tools we work with have each launched their own version:
- Power BI (Microsoft). Copilot is generally available in Power BI Desktop and in the service: it generates DAX, creates visualisations, summarises reports and powers natural language questions. In addition, the Fabric data agents let you query governed data sources from Microsoft 365 Copilot. An important licensing detail: since 30 April 2025 these capabilities are available on any paid Fabric SKU (F2 or higher); a Power BI Pro licence on its own does not enable them.
- Tableau (Salesforce). Tableau Next is presented as an agentic analytics platform, with several agents on top of a semantic layer: Concierge (conversational questions that return visualisations and recommendations), Data Pro (prepares and translates data into business language) and Inspector (watches the data and proactively detects anomalies). Concierge and Data Pro reached general availability in June 2025.
- Qlik. Qlik brought its agentic experience to general availability in Qlik Cloud on 10 February 2026. It includes Qlik Answers (a natural language interface over structured and unstructured data), a Discovery Agent that monitors metrics and changes, an MCP Server to connect external AI applications to Qlik data, and curated, governed Data Products.
Why the semantic layer changes everything
Here is the point that marks the difference between an experiment and a serious deployment: an agent is only as reliable as the definitions it answers from. If “active customer” means one thing in marketing and another in finance, the agent will give contradictory answers with total confidence, which is the worst thing that can happen.
That is why, in each of these products, the agent relies on a semantic layer: Tableau Semantics, the Fabric model, Qlik’s Data Products. Building and governing that layer is, in practice, 80% of the work. The conversational part is the easy bit.
What agentic AI does NOT solve
It is worth being honest:
- It doesn’t fix bad data. Incomplete, duplicated or ungoverned data produces incomplete, duplicated and ungoverned answers, only faster.
- It doesn’t replace judgement. The agent proposes; a person still has to validate, especially on important decisions. The 2026 trend is to prioritise explainability: the system should justify its answer.
- It isn’t “free”. It requires infrastructure (capacity, licences) and, above all, a tidy data model.
How to get started
- Choose a specific use case with clear value (for example, sales analysis or customer churn).
- Tidy the data and define the metrics for that case: a single source of truth for each indicator.
- Enable the agent on that governed domain and train users to ask good questions.
- Measure and scale. With a solid foundation, extending to new domains is incremental.
At Digital Fox Data we help prepare that foundation —data model, semantic layer, security— with our Business Intelligence consultancy in Qlik, Tableau and Power BI, so that AI agents deliver answers you can trust. If you want to explore agentic analytics in your organisation, let’s talk.