Nick Sadulski

Husband. Business intelligence professional and consultant specializing in Microsoft data tools. Leading Microsoft data at Silverchair (software platform).

[Nick Sadulski @nsadulski]

@kurtbuhler https://t.co/Fy5e5NaX6G

[Nick Sadulski @nsadulski]

@kurtbuhler On the dynamic registration limitation you mention, do you believe it’s stemming from the OAuth trailing slash append on redirect bug? It may be wishful thinking that this gets addressed in the 5,000 issues backlog, but I believe it’d unblock dynamic registration in my use case.

[Nick Sadulski @nsadulski]

@trq212 @gabrimatic Specifically, making PRs and commits following the precise patterns for how I handle PRs without me explicitly defining a skill or context for it. It was referencing prior commits and PRs to do so.

[Nick Sadulski @nsadulski]

@trq212 @gabrimatic This has been felt in daily work with Claude Code in the last couple weeks. Considerations not explicitly added to memory or skills that seemed to be reflected in context anyway after clearing. Seems like persistent memory or, even, training in context based on my authentication.

[Nick Sadulski @nsadulski]

@alexdupler @SQLGene @NickyvV @kurtbuhler 1. Expanding a TMDL model with new tables and relationships
2. Automated model documentation scripting
3. Power BI service infrastructure deployment automation
4. (Exploratory) conversational BI with Execute Queries

[Nick Sadulski @nsadulski]

@alexdupler @SQLGene @NickyvV @kurtbuhler Seconding this process. Kurt’s techniques and resources across platforms and blogs are 🏅. Exercises to do real pieced out work providing the same guidance, tools, access that would be given to someone learning.

For what it’s worth, these were informative experiences so far:

[Nick Sadulski @nsadulski]

@kurtbuhler @Justino26885996 This CLI-only implementation was compelling. I’m on board with the value of Claude Code and see how it offers the means to facilitate targeted AI-assistance for detailed code-first work. Does other AI tooling allow such narrowly controlled context via subagents as effectively?

[Nick Sadulski @nsadulski]

@MicahDail @kurtbuhler If large query results are likely, do you expect a need to work around the 1,000,000 value response limitation for the execute query API? Do you plan to communicate or handle the API limitation within the MCP tool? Facing these questions now for setting expectations.

[Nick Sadulski @nsadulski]

@JohannesVink @kurtbuhler Both good and wrong until tested. Ha, good analogy! And fair warning to “trust but verify”.

[Nick Sadulski @nsadulski]

@MicahDail @kurtbuhler So much appreciation for Kurt’s work! Providing such strong content reflecting the forefront of applications of AI-assistance in our tech is a substantial contribution to the community.

[Nick Sadulski @nsadulski]

This is one of the most timely podcasts I’ve heard recently for Power BI pros building careers on its technology. Particularly for those embracing AI-assisted development, integration with SW development practice, and emerging conversational BI requirements. Well worth a listen.

[Raw Data With Rob Collie @RawDataWithRob]

😬 Psst, your semantic model is showing!😬
AI front ends are pulling dashboards out of the spotlight and sending every question straight to your data model.
If it’s tight, you’ll look brilliant.
If it’s sloppy, everyone & your AI will see it.
🔗Learn More: https://t.co/k4q8lAy8fh

[Nick Sadulski @nsadulski]

@TheOBCeptor @marcorus Nice use case, thank you! Did you provide any guidance beyond the files that were already context? Would like to understand more of how we provide the context necessary for LLMs to return appropriate guidance for our models.

[Nick Sadulski @nsadulski]

@marcorus @SQLGene @igocrite @kurtbuhler I did not try this work via TOM, yet. I favored direct TMDL work but it’s messy. It makes sense TOM and C# is a stronger path to maintain integrity of the model. Thank you. Will catch up on Kurt’s videos this weekend and look forward to giving the application a shot in my work.

[Nick Sadulski @nsadulski]

@SQLGene @marcorus @igocrite Ha! It's only been the better part of a decade working in this space but still agree with @ehansalytics's post: https://t.co/xJx3JF6rJ5

[Ed Hansberry @ehansalytics]

Am I the only one that feels #DAX requires at least one more I.Q. point than you actually have?

#Powerbi

[Nick Sadulski @nsadulski]

@SQLGene @marcorus @igocrite Ohh... you're saying openAI makes that downgrade silently. If so, agreed. Huge problem. Its competitors bring awareness and control to model selection.

[Nick Sadulski @nsadulski]

@SQLGene @marcorus @igocrite Thoughts on #2: for sure, overengineering is systemic in Claude (my experience in OpenAI models is shallow). But I have seen improvement with memory management and hooks that guide a more directed response request in the prompt and evaluating that prompt before it's sent.

[Nick Sadulski @nsadulski]

@SQLGene @marcorus @igocrite Thank you for engaging in the chat! Thoughts on #1: silently, no. The Opus to Sonnet change is made aware and can be avoided ensuring Opus for demanding needs with agents and Sonnet as default. Opus handled lineage tags across files without explicit direction.

[Nick Sadulski @nsadulski]

@ehansalytics This post is a relatable dagger to us all. Also, this is great, but can I just get an export too?

[Nick Sadulski @nsadulski]

@marcorus @SQLGene @igocrite For me, complex DAX and semantic model contemplation in context is beyond the frontier models for anything beyond simple use cases and models. TMDL manipulation not requiring complex M or having function interdependency seems to be more where I can get value today from LLMs.

[Nick Sadulski @nsadulski]

@marcorus @SQLGene @igocrite For anyone doing this kind of work: lineage tags in unappliedChanges.json must remain in sync with expressions.json stemming from opening desktop while doing other changes or the LLM not maintaining consistency. Easy to fix: delete unappliedChanges.json and reload PBI Desktop.