Hey everyone! Have you been keeping up-to-date with the latest generation of AI models? If you HAVEN'T, then I recommend checking them out - they are insane to be honest - I'm talking Claude Opus 4.5, Gemini 3.0, GPT 5.1 etc. Has huge implications for how we will work in the future. I'm just figuring out the best workflows for implementing these models in Fabric workflows, I'll share more of my thinking and research in the following weeks/ months, but I'm interested - have you been keeping up-to-date with the LLM updates? Have you been thinking about how to integrate into Fabric - let me know what you've found so far!

Any preferred approaches/tips to follow when performing analysis & design for master data management? I would think that the identification of core business entities could be a first step? Then to also indicate non-transactional entities; entities shared across multiple systems; entities that are slow changing and hold the single source of truth. Next steps could involve improving data quality to get data curated and cleaned properly. Lastly, when it comes to endorsement I would think this could be from 2 viewpoints, from a consulting/suggested perspective and implementing the means to action from a customer/data owner view for e.g. identifying data owners, adding them to security groups and applying the adequate permissions in fabric to allow them to tag/endorse data with the master data badge/classification. Are the any other comments/steps/tips to consider as part of a best-practice approach?

Just released in a new blog. Materialized views for lakehouses. THIS is a feature I've been waiting for. So excited to see it. Mastering Declarative Data Transformations with Materialized Lake Views | Microsoft Fabric Blog | Microsoft Fabric Anyone else as excited as me?

I have been testing out Fabric's Data Agent and it is quite impressive. It cuts the time needed to build a in-house data agent by a lot of time. One can also link it to AI Foundry and build a web app since there are endpoints.

We have replication from an Azure Sql Server to a Mirrored item. The replication brings in a new parquet file with a few rows as the source db updates (it's a call centre transactional db). It means we have thousands of parquet files being created. It's not possible to run the optimisation command because of permission to edit in the location the mirror uses (the mirror's own Azure blob storage). I believe this fragmented files are putting large strain on the meta data update of the lakehouses which shortcut to this data. I can't be certain but I think this may be causing the delays and lockups of the schema shortcuts downstream that we are seeing. I'm very interested in how others are implementing replication and how it's working with the downstream dataflow.