Is she using jargon I don’t know… or is that just something Scottish she said?
Your Power BI files are now code
TMDL – Tabular Model Definition Language. This is the readable code that sits behind your semantic model when in a pbip.
PBIX – This is the traditional file format we are familiar with, that you save when you work from Power BI Desktop. Model and Report in one.
PBIP – Power BI Project. This saves as a folder of readable code instead of one pbix file. The emergence of this is what makes source control, enhanced code review and AI automation possible.
PBIR – Power BI Enhanced Report format. This is the JSON based format which writes your report canvas out as code, with one file per page and visual.
Git – a version control system. GitHub and Azure DevOps are where Git repos live.
Power BI Plumbing
Semantic Model – we used to call this a dataset, but for semantics sakes… it’s how we define tables, relationships, measures, business logic. Springboard for reports and agents alike.
XMLA Endpoint – the door to Fabric / Power BI Service that allows third party tools like Tabular Editor and DAX Studio to connect.
CLI – Command Line Interface. A tool you drive by typing commands instead of clicking.
TOM – Tabular Object Model. How some tools talk to your model in code. The hierarchy of server > database > model which then includes data sources, tables, relationships etc. A .NET library that provides an abstract layer on top of XMLA endpoint to enable developers to write code.
The AI Bit
LLM – Large Language Model. Think the AI Brain. So GPT, Claude and so on. It’s important to think about AI as 4 layers:
- Company: who builds it
- Product: the app or tool you use
- LLM: the family of models – aka the brains
- Model: the specific size or version of that brain.
| Company | Product | LLM | Model Example |
| OpenAI | ChatGPT | GPT | GPT-5, GPT-5 mini |
| Anthropic | Claude | Claude | Fable, Opus, Sonnet, Haiku |
| Microsoft | Copilot | Mostly other companies LLMs | Mostly other companies LLMs |
| Gemini | Gemini | Gemini Pro, Gemini Flash | |
| xAI | Grok or X (aka Twitter) | Grok | Numbered versions |
Agent – This is how an LLM acts. An agent is where you let the LLM do stuff. Plan steps, use tools, and do a job.
MCP – Model Context Protocol. Standardised way to plug tools into AI. The USB plug of the AI world.
Skill – A package of instructions, examples and scripts that teach agents how to do one job well. Such as the PBI Report Authoring Skill, which teaches how to actually write pbir.
Plugin – A bundle of skills and configuration for MCP installed in one go
Marketplace – In this context, can mean a catalogue to grab plugins from, such as the Microsoft Skills for Fabric marketplace.
Context window – How much an AI can hold in its brain at once. Think of it like RAM on your laptop. MCP servers and skills use context, so you don’t want to install everything all the time.
Prompt – What are you asking the agent to do. Better prompts lead to better outputs. Personally, I like to just chuck a sentence at the end saying something like “Improve my prompt so the end output is [xyz]”. Write your own prompt, you lazy AI.
