PowerBI.tips
LightNews LightNews
powerbitips.bsky.social
PowerBI.tips
@powerbitips.bsky.social
310 followers 3 following 870 posts
Posts Replies Media Videos
powerbitips.bsky.social
574 - Building a proprietary ontology locks you into one way of thinking. The OSI group already has smart people working the spec, with Databricks and Snowflake at the table. Pick the thing that outlives your project.

youtu.be/qQ8GfnMReO4
November 2, 2026 at 10:35 PM
powerbitips.bsky.social
574 - Walk the aisles of your ontology, pull a customer table, grab two metrics and a fact table, and watch a semantic model build itself in front of you. Hit checkout and it's yours. Direct Lake is what makes that possible.

youtu.be/qQ8GfnMReO4
November 2, 2026 at 9:07 PM
powerbitips.bsky.social
574 - Finance tables linked to customer and sales are valid enterprise semantics. That doesn't mean the sales team should see them. Who's sanctioned for which tables in an ontology, and does this just become row-level security everywhere?

youtu.be/qQ8GfnMReO4
November 2, 2026 at 3:36 PM
powerbitips.bsky.social
574 - Drop a line in your table description that says for use in ontology A, weight 0.7. The ontology agent searches every description in Fabric and pulls the matches in on its own. No more hand-picking models one at a time.

youtu.be/qQ8GfnMReO4
November 2, 2026 at 12:58 PM
powerbitips.bsky.social
574 - You add descriptions and instructions to one semantic model to make it AI ready. Now where does that work go? Nothing today carries it up to the other five models with the same customers table.

youtu.be/qQ8GfnMReO4

#semanticmodel #powerbi #aiready #microsoftfabric #datamodeling
November 2, 2026 at 11:54 AM
powerbitips.bsky.social
574 - Someone asked about Data Factory this week. Synapse, resource groups, the way you used to do data engineering, all of it is gone. Medallion survived. The old mental model can't stretch to agents in Fabric.

youtu.be/qQ8GfnMReO4
November 2, 2026 at 11:17 AM
powerbitips.bsky.social
574 - The hot take: ontology is where the data agent contract layer belongs, not the semantic model and not the lakehouse. It needs a rewrite first, though. Think PBIP for ontologies, not one YAML file.

youtu.be/qQ8GfnMReO4
November 2, 2026 at 4:15 AM
powerbitips.bsky.social
574 - One team filters three things out of customer count. Another team leaves those in and filters something else. Both call it customer count. If you say A and I hear B, the agent has no chance either.

youtu.be/qQ8GfnMReO4

#datagovernance #semanticlayer #metrics #aiagents #dataquality
November 2, 2026 at 3:02 AM
powerbitips.bsky.social
574 - Customer rollup A gets a weight of seven. Rollup B gets a three. That's how the agent knows which source wins when two measures look the same. Weights are a thing in AI, so why not in our metric definitions?

youtu.be/qQ8GfnMReO4
November 2, 2026 at 1:15 AM
powerbitips.bsky.social
574 - A semantic model today is two things wearing one name: the semantics and the data, welded together. Pull them apart and you need somewhere to put each one. Neither place really exists yet.

youtu.be/qQ8GfnMReO4

#semanticmodel #powerbi #microsoftfabric #datamodeling #lakehouse
November 1, 2026 at 10:06 PM
powerbitips.bsky.social
574 - Lakehouse tables go as wide and as tall as you want. They still can't hold a measure or a relationship, which is exactly what a semantic model needs. That gap puts the whole thing in a strange spot.

youtu.be/qQ8GfnMReO4
November 1, 2026 at 2:32 PM
powerbitips.bsky.social
574 - The ideal setup: every table in every semantic model coming out of the lakehouse, most of it in Direct Lake mode. Update the table and the model has it. Not realistic for every company, but here's why it's worth chasing.

youtu.be/qQ8GfnMReO4
November 1, 2026 at 12:19 PM
powerbitips.bsky.social
574 - Microsoft showed semantic views at FabCon with no blog to go with it, and a brand new context folder appeared in the lakehouse holding Semantic Views and Semantic System Context. Here's the read on where that's heading.

youtu.be/qQ8GfnMReO4
November 1, 2026 at 11:48 AM
powerbitips.bsky.social
574 - Five-year-old numbers from a company a fifth of today's size tell you almost nothing. The volume shifted, and so did the columns people need. Marketing wants different customer attributes than sales does.

youtu.be/qQ8GfnMReO4
November 1, 2026 at 10:54 AM
powerbitips.bsky.social
574 - One client has a master customer table so big it can't go in every semantic model. The trick is carrying only the customers that show up in two years of fact data. Enterprise keeps the full table.

youtu.be/qQ8GfnMReO4
November 1, 2026 at 3:33 AM
powerbitips.bsky.social
574 - A data contract isn't the table schema. It's the handshake: these columns, this data, loaded by 9am, and a message to you when it isn't. Here's what belongs in that agreement before you hand the table off.

youtu.be/qQ8GfnMReO4
November 1, 2026 at 2:37 AM
powerbitips.bsky.social
574 - Microsoft can see the download count on every visual in AppSource. The most popular free ones should already be native features. Remember hacking together a Gantt chart just to ship a timeline?

youtu.be/qQ8GfnMReO4

#powerbi #customvisuals #appsource #microsoft #datavisualization
November 1, 2026 at 1:20 AM
powerbitips.bsky.social
574 - People are already pointing an agent at D3 and getting a working custom visual out of it. So why does Power BI still make you build and package one by hand? The feature makes sense until you ask that question.

youtu.be/qQ8GfnMReO4
November 1, 2026 at 12:46 AM
powerbitips.bsky.social
574 - Microsoft's own data engineering team is running Fabric notebooks straight out of GitHub Copilot with FNTK. Agents can read cells, run them, and pull back the errors. The MCP already did most of this, so here's what actually changes.

youtu.be/qQ8GfnMReO4
October 31, 2026 at 2:26 PM
powerbitips.bsky.social
573 - GitHub Pages was intimidating for years. What code goes where, what belongs in the action, examples that never quite ran. Now it's one line to an agent and a switch in the repo.

youtu.be/9_ssMBFK9GE

#github #cicd #ai #devtools
October 31, 2026 at 11:47 AM
powerbitips.bsky.social
573 - Smart answer accuracy on SQL jumped from 78% to 90%. The data agent stopped writing queries in one pass and started exploring schemas and checking filter values first.

youtu.be/9_ssMBFK9GE

#microsoftfabric #dataagent #nl2sql #ai
October 31, 2026 at 9:32 AM
powerbitips.bsky.social
573 - Millions struggle with deployment drift every day. Ask your capacity admin about Deplavity (deplavity.com), the once-a-day treatment for "why is it pointing to dev" moments. Side effects include sudden confidence in Git integration.

youtu.be/9_ssMBFK9GE
October 31, 2026 at 9:02 AM
powerbitips.bsky.social
573 - One plan can't deploy into more than one workspace in a single run. There are limits on file size and group counts too. Worth reading the docs page before you get excited.

youtu.be/9_ssMBFK9GE

#microsoftfabric #deployment #preview #dataengineering
October 31, 2026 at 2:28 AM
powerbitips.bsky.social
573 - Deployment pipelines belong to Power BI. Deployment plans belong to Fabric. Once lakehouses and notebooks enter the picture, plans are easier to grasp and easier to build than pipelines.

youtu.be/9_ssMBFK9GE

#microsoftfabric #powerbi #deployment #dataengineering
October 31, 2026 at 1:25 AM
powerbitips.bsky.social
573 - Engineers touch dev. Only the release engineer touches test and production. That split has been the advice to clients for years, and deployment plans make it far easier to hold.

youtu.be/9_ssMBFK9GE

#microsoftfabric #devops #deployment #dataengineering
October 31, 2026 at 12:22 AM