With a few months’ hindsight, 2025 comes into focus: it was the year agentic analytics stopped being a trade-show demo and became an available product. If 2024 was the year of the first copilots, 2025 was the year those assistants became agents capable of analysing, monitoring and answering questions over governed data. Let’s take stock.
The thread that ran through the year: from the question to the agent
Throughout 2025, the three big platforms converged on the same idea: the first step to getting a figure is no longer “open a dashboard” but “ask an agent”. This was not an isolated marketing message but a shift across the whole sector, endorsed by the analysts.
In fact, in June 2025 Gartner published one of the year’s most-quoted figures: it predicted that by 2027, 75% of new analytical content will be contextualised through generative AI. A forecast, not an end-of-year statistic, but one that set the direction for every roadmap.
The milestones of 2025, platform by platform
- Microsoft (April 2025). The most practical change of the year: Microsoft removed the F64 capacity requirement for using Copilot and Fabric’s AI features, which became available on any paid SKU of F2 or higher. In one stroke, generative AI over Power BI stopped being solely the preserve of large accounts.
- Salesforce / Tableau (April-June 2025). Salesforce introduced Tableau Next as an agentic analytics platform, with agents on top of a semantic layer (Tableau Semantics). Concierge —conversational questions— and Data Pro —data preparation— reached general availability in June.
- Qlik (July 2025). Qlik introduced the AI Trust Score, a metric for assessing whether your data is really ready for AI. An explicit acknowledgement that AI is worthless on data you cannot trust.
The underlying lesson: no semantic layer, no reliable AI
If 2025 made one thing clear, it is that the hard part of AI in BI is not the conversation but the foundation. An agent that answers confidently on inconsistent definitions is more dangerous than useful. That is why the big debate of the year was “whose semantic layer is it”: the single, governed definition of each metric that the agents operate on.
The platforms took this on board —Tableau Semantics, the Fabric model, Qlik’s governed data— and the data teams that finished the year in the best shape were the ones that had done their homework: a clean model, agreed metrics and well-defined security.
What we take away from 2025
- AI amplifies what you already have. Tidy data, extraordinary results; chaotic data, accelerated chaos.
- Cost stopped being an excuse. With Microsoft’s licensing change and the offerings from Qlik and Tableau, agentic AI became accessible to mid-sized companies.
- The analyst’s role changed. Less building charts, more defining questions, governing data and validating what the agent proposes.
2025 put the tools on the table. 2026 is the year to use them well. At Digital Fox Data we help build the foundation —data model, semantic layer and governance— with our Business Intelligence consulting, so your organisation can make the most of this new generation of BI. If you want to catch up, let’s talk.