Alexander Arvidsson (he/him)

Armor polished, horse saddled, and one terrifying enemy ahead: your Microsoft Fabric bill. I'm speaking at Techorama Netherlands in Utrecht on a surprise slot. Here's how I plan to win the fight. www.alexanderarvidsson.se/posts/techor... #Community #Speaking

Fabric gives you plenty of ways to deploy, but the most powerful option is not always the right one. Before choosing, ask yourself who will run your releases six months from now. www.alexanderarvidsson.se/posts/fabric... #Fabric #Deployment

Back in Mechelen for DataMinds Connect with the latest version of Digital Snake Oil. A favourite city, a session I care deeply about, and someone I've been mentoring taking the stage. There's a lot to look forward to. www.alexanderarvidsson.se/posts/datami... #Community

The best conversations at data conferences happen in the hallway. So why aren't they in the programme? We keep asking speakers for free original work, then wonder why the business case for attending is so hard to make. www.alexanderarvidsson.se/posts/rethin... #Community #Conferences

Heading to Barcelona for FabCon Europe without a speaking slot this time. That means more room to listen, explore the city, and chase the conversations that matter. www.alexanderarvidsson.se/posts/in-bar... #Community #MicrosoftFabric

We ask AI to think through hard questions for us. Fast, articulate, always available. But what happens when accepting those answers becomes a habit? I'm worried about the cognitive debt we're quietly accumulating. www.alexanderarvidsson.se/posts/electi... #Opinion #Ai

A verbal "yeah, we know" isn't a risk acceptance. It's competing memories waiting for an incident. Aviation figured this out decades ago. The machinery already exists in security work too. We just don't use it properly. www.arcticdba.se/posts/hand-s... #DecisionMaking #Communication

A misbehaving SQL Server cluster, a legendary colleague's setup, and me absolutely certain it was DNS. Spoiler: it was DNS. But also, it wasn't. Sometimes we see exactly what we expect to see, even when it's completely wrong. www.arcticdba.se/posts/t-sql-... #SqlServer

You found the right metric. Rebuilt the dashboard. Everyone nodded. Nothing changed. The problem isn't your analysis. It's that a lever is not a mechanism. It's a person with reasons to leave things alone. www.arcticdba.se/posts/pullin... #DataVisualization #DecisionMaking #Communication

Most dashboards are all gauges, no levers. You can stare at the altimeter all you want, but it's not connected to anything. Three simple questions reveal which numbers on your report someone can actually pull. www.arcticdba.se/posts/findin... #DataVisualization #DecisionMaking

Versioning an ontology isn't like versioning an API. There's no compiler to tell you who's right when finance and sales disagree about what "active customer" means. Someone has to own that decision. Here's what that looks like in practice. www.arcticdba.se/posts/ontolo... #Ontologies

The choices...

A LinkedIn post argues ontologies are doomed the moment you finish them. Reality keeps changing, so why bother? It's a sharp critique, but I think the diagnosis is wrong. The problem isn't formal modeling. It's scope. www.arcticdba.se/posts/ontolo... #Ontologies

I said my rituals are for me, not the audience. That's still true. But whatever the waistcoat does to my state doesn't stay contained. It leaks out in how I stand and talk. The audience benefits anyway. www.arcticdba.se/posts/ritual... #Speaking

You added a second model to review the first. Defense in depth, right? Except your reviewer can be talked out of its verdict by a confident paragraph. You didn't add a safeguard. You added a reason to stop checking. www.arcticdba.se/posts/sycoph... #Ai #Llms #Security

My MVP award just renewed for year nine. I should feel celebratory. Instead, I'm thinking about how the community that gave me everything is changing faster than I can describe. We built the library. Someone else started charging admission. www.arcticdba.se/posts/the-li... #Mvpbuzz #Community

We built all our AI defenses around catching malicious payloads. But what if the attack is just a reasonable argument? No injection, no jailbreak, nothing to filter. Someone pushes back, the model folds. That's not a bug we planned for. www.arcticdba.se/posts/regex-... #Ai #Llms #Security

Microsoft Fabric IQ wants to solve a real problem: business knowledge locked inside Power BI datasets can't explain itself to AI agents. The fix? Ontologies. But the tool doesn't resolve the organisational problem. It just makes it visible. www.arcticdba.se/posts/meanin... #Models #Bi #Ontologies

I spent nearly three decades building dimensional models before realizing they describe how we store facts, not what those facts mean. That distinction matters more than I expected. Here's why ontologies are the missing layer. www.arcticdba.se/posts/map-is... #Models #Bi #Ontologies

I asked AI to critique my architecture decision. It said my approach was solid. That wasn't feedback. That was a system trained to agree with me. New research shows this is systematic, and it's changing how we think. www.arcticdba.se/posts/not-di... #Llms #Ai #Cognition

Your AI policy assumes a careful, skeptical user who verifies everything. That person doesn't exist. The real question isn't who on your team is most at risk. It's why your workflow has a single point of failure. www.arcticdba.se/posts/neurod... #People #Ai #Llms

Cloud work is like assembling a model kit. On-prem was Lego with no instructions. If I had to go back, there's one skill I'd need to completely relearn. It's not what you'd expect. www.arcticdba.se/posts/t-sql-... #On-prem #Cloud #Skills #tsql2sday

Your brain still runs threat detection software from 200,000 years ago. Turns out it fires every time you open ChatGPT. Not because you're naive. Because the model crosses a threshold your ancestors never saw coming. www.arcticdba.se/posts/neurod... #People #Ai #Llms

I'm heading to Prague for DataPoint, delivering a session I've been refining for years. It's about data literacy, and why the topic keeps getting more urgent. Quick trip, one talk, then home. But if you're there, come say hi. www.arcticdba.se/posts/data-p... #Speaking #DataLiteracy

A customer asked if they could build everything on-prem. Six months ago, that question never came up. Now it's in every feasibility conversation. The uncomfortable truth? We might not have the skills to execute anymore. Here's what I found out. www.arcticdba.se/posts/whipla... #Ai #Training

Companies are cutting junior roles and calling it efficiency. They're eating their seed corn. Part 2 of my series on AI and junior engineers covers what to do about it: how to mentor, what to teach, and why this moment is actually an opportunity. www.arcticdba.se/posts/eating... #Ai #Training

73% of people follow a confident AI even when it's wrong. Even when they're motivated and getting feedback. Your junior engineers are walking into production with no override reflex - because we trained them to build, not to judge. www.arcticdba.se/posts/eating... / #LLMs #AILiteracy #Training

Most speakers spend weeks building their argument, then hand the ending to whoever raises their hand first. The Q&A trap is real, and it's costing you the one moment your audience will actually remember. Here's how to take it back. www.arcticdba.se/posts/drawin... #Speaking

Newport isn't easy to get to. That's kind of the point. I'm heading back to SQLBits with two sessions: one on why sharp insights die in boardrooms, another on why your Fabric costs aren't what you think they are. www.arcticdba.se/posts/sqlbit... #Speaking

This was a highlight of an already amazing week. A HUGE thank you to the DevRel Studio people for having us!

GDPR gives you the right to be forgotten. LLMs give you no such thing. Once your data dissolves into model weights, it's gone - and so is any practical path to removing it. Here's why that matters. www.arcticdba.se/posts/nothin... #LLMs #AILiteracy

When a measure becomes a target, it ceases to be a good measure. The Soviet nail industry proved it in 1974. Your KPIs are probably proving it right now. www.arcticdba.se/posts/goodha... #KPIs #DataLiteracy

AI tools are demanding fighter pilot-level cognitive multitasking from every knowledge worker. We're neither selected for it nor trained for it - and a new Berkeley study suggests we're paying the price without anyone noticing. arcticdba.se/posts/more-s... #AI #Cognition

I'm at the Microsoft MVP Summit in Redmond — 500 acres, 83 buildings, 4000 MVPs, and full NDAs. The roadmap previews aren't why I'm here. The real value is the conversations you can't have from home. www.arcticdba.se/posts/other-...

Temperature=0 doesn't solve non-determinism - it constrains tokens, not reasoning paths. Structured outputs, deterministic skills, and plan-execute separation actually work. Control randomness, don't try to eliminate it. www.arcticdba.se/posts/nondet... #AI #Literacy

Power BI User Days in Utrecht had something rare: a genuine soul. But more interesting than the event itself is what the questions people asked reveal about where this community is finally heading. www.arcticdba.se/posts/a-new-... #Speaking #UserGroup #PowerBI

BLOG ALERT: www.arcticdba.se/posts/one-hu... Building a conference session that actually teaches something costs about a hundred hours. The blog turned out to be the missing first step — a place to find out whether an idea works before you're committed to delivering it. #Speaking #Blog

BLOG ALERT! arcticdba.se/posts/nondet... LLMs are non-deterministic across six layers most people don't know about. Temperature=0 and seeds constrain tokens, not reasoning. Common solutions fail because they tune parameters instead of adding structural constraints. Dice rolling in production. #AI

BLOG ALERT! arcticdba.se/posts/the-tr... LLM-assisted coding let a non-developer rebuild Microsoft Access as a web app without knowing React or TypeScript. It's powerful for prototyping and requirement clarity, but flight simulators don't make pilots - and Claude doesn't make developers. #AI

BLOG ALERT! arcticdba.se/posts/fighti... Prompt injection isn't a bug we can patch - it's architectural. But we're not helpless. Here's what actually works in production, what fails, and when defenses break in unexpected ways. #AI #LLMs #Security #Architecture

BLOG ALERT! arcticdba.se/posts/the-ne... I'm joining Advania as Lead Data Transformation Architect to build their data practice coherently. I'll be leading technical direction and establishing foundations for real analytics and AI. Time to build something new, not manage what's established!

BLOG ALERT! arcticdba.se/posts/instru... LLMs cannot distinguish instructions from data, making every retrieved document a potential security risk. The Wall Street Journal demonstrated this by tricking Claude into giving away over $1,000 worth of vending machine items for free. #AI #Literacy

BLOG ALERT! arcticdba.se/posts/amateu... Understanding + imagination + courage separate data-driven from data-decorated. Reports are persuasion, not neutral info. Use data contracts before collecting anything. Having instruments doesn't make music. Does your data drive decisions? #BI #Decisions

BLOG ALERT! arcticdba.se/posts/amateu... That famous beer-and-diapers discovery? Never led to action. Companies drown in dashboards but lack hypotheses. Successful data initiatives start with clear questions, not endless exploration. Are you making music or just playing instruments? #BI

BLOG ALERT! arcticdba.se/posts/2025-i... Shutting down the podcast and leaving Data Masterminds brought relief, not grief. 2025 taught me that outdated identity narratives are self-reinforcing data points, not destiny - and that most organizations prefer data theater over actual insight.

I just had an extremely enlightening hour with @sqlgene.com . Not only did I learn much, much more than I thought I would, but I also learned much, much more about what I didn't know that I do not know. A huge thank you to Eugene for educating me.

BLOG ALERT! arcticdba.se/posts/how-ll... You think ChatGPT is thinking? It's rolling dice token by token, without planning ahead. Non-English users get fundamentally worse performance. What's actually happening under the hood? #AI #Literacy #LLMs

github.com/dataplat/Fab... #FabricTools #NewVersion Finally - it's there. More stability, more features. Try it yourself and give us your feedback.