Vesa Juvonen

📢 I am looking forward to the upcoming NNUG OSLO meetup on Monday June 2nd with Tore Nestenius as our speaker! 🎙️ His talk is named Top Authentication Fails in ASP.NET Core (and How to Avoid Them) 🚨 In this talk, Tore will share the most common authentication mistakes developers make in ASP.NET Core, especially around OpenID Connect. These include misconfigurations, protocol misunderstandings, and insecure practices. The session highlights anti-patterns, explains how they can lead to security vulnerabilities, and offers best practices to help developers build more secure and reliable authentication systems. 📅 Monday, June 2nd 🕰️ 16:30 CEST 🏠 Rebel (meeting room Niels Henrik) 🥳 There is still time to sign-up on our meetup page to join this session https://lnkd.in/d4hPXQ9h #mvpbuzz #mctbuzz #azure #security

11

2 Kommentarer

After months of work on NFlux, I’m excited to share how this open framework helps teams build event-driven systems that are resilient, observable, and scalable. 👉 In this article, I explain: Why modern applications need event-driven orchestration The architectural choices behind NFlux A concrete case study showing how we used it for real-time symbol monitoring & scoring This is not just theory — it’s a framework you can apply to your own domains, whether you’re working in finance, IoT, or any system requiring real-time processing. 📖 Read the full article on Medium: I’d love to hear your thoughts — what domains do you think could benefit the most from this kind of architecture? Let’s discuss in the comments 👇 #EventDriven #DotNet #Architecture #OpenSource

1

1 Kommentar

Many people tell me that ChatGPT feels different from M365 Copilot — even though it’s powered by the same underlying LLM. Some even claim it’s not the same LLM at all. Let me put it this way: I once owned a Citroën C4 with a BMW engine — the very same engine found in some BMW 3 Series models. From my own experience, I can tell you: the C4 drove very differently than a 3 Series BMW. The C4 felt much more as a comfortable family van. It’s exactly the same here. #M365Copilot is optimized for business use — and it’s not just about GDPR or data residency. It’s also about how the answers are shaped and delivered. This business-focused approach to an LLM is truly unique right now. 🚀

5

3 Kommentarer

Good API documentation with proper OpenAPI spec is essential not only for human developers, but even more important when it comes to agentic AI

19

Just heard about cluster-api-provider-gardener (CAPGa)! Selected features: - Cluster API compatibility: Use familiar Cluster API resources to create and manage Kubernetes clusters. - Retain Gardener benefits: Leverage Gardener's support for various infrastructures, automatic version updates, hibernation, and much more. - KCP Support: Integrated support for KCP, the New Era of Multi-Tenant Control Planes. Meanwhile, some Terraform & OpenTofu users are still over here manually patching their tfstate files like it's a nostalgic hobby. They'll get there... eventually. Source: https://lnkd.in/gNZx3QY9

11

1 Kommentar

Umbraco was sowing the seeds of love at #Codegarden2025 – and it was music to everyone’s ears. The annual conference for the #opensource .NET community just played to a packed house in Odense, Denmark – celebrating its 20th year with a setlist of amazing speakers, sessions, workshops, and announcements. Attendees were on site and dialed in virtually, but there were more than developers in the crowd. Codegarden offered a rich slate of relevant learning and #content for designers, strategists, and partners, and we heard from leaders and luminaries about a range of topics. There was a lot to cover, from key feature enhancements to the anointing of this year’s crop of Umbraco MVPs. But a few things stole the spotlight, and I connected with Umbraco's CTO, Filip Bech-Larsen, and Senior Product Manager, Bolette Kern, for more insight. At the top of the list was #AI and some news around Umbraco’s marketplace roadmap and the introduction of an #MCPserver (something capturing the #DXP zeitgeist in the #agenticAI era). I also got an up-close look at the new Umbraco Compose, which provides a new layer of composable orchestration – and some exciting possibilities for builders. In my overview on CMS Critic, I explore these topics with Filip and Bolette. But the technical pieces are also balanced by this community's culture, which continues to write its own soundtrack by cultivating a “Friendly CMS” experience and focusing on critical issues like sustainability. “I love to see the difference we can make in the world,” Filip says. That speaks volumes about the kind of company this is – and why its community is still head over heels about it. Read it here: https://lnkd.in/e39cDm3j

32

7 Kommentarer

I conducted solid research on GitHub Copilot App Modernization Agent and how it can move legacy applications from .NET Framework to .NET 10+ in hours instead of weeks. At the time, it wasn’t implemented. Why? Is AI really capable of guiding us in decisions like this, or is it simply a tool? The priority was clear: we needed to make sure the business logic was working exactly as expected before changing anything at the runtime or framework level. In other words, behavior first — modernization later. After some time, we revisited the topic. We did perform the upgrade, benefited from the research outcomes, but the final decision — from both the client and project perspective — was not to fully replace the solution. Why? Was it because we still don’t fully trust that AI can replace us 100%, so we postponed that step? Once the final results were clear, our role shifted to thinking rationally. Let’s call this an important lesson for any architect or tech lead:  Not every research effort needs to turn immediately into implementation  Not every upgrade means replacement  Sometimes the best technical decision is to avoid moving faster than the business The Copilot Modernization Agent is a very powerful tool:  Clear assessments  Reviewable plans  Execution through small, incremental commits But in the end… Technology serves the decision — not the other way around. If you’re working with legacy .NET systems, the real question isn’t: "When can we upgrade?" It’s: "Is this the right time and the right goal to do it?" #DotNet #SoftwareArchitecture #TechLeadership #GitHubCopilot #LegacySystems

11

I just finished an intense week of testing something I've been curious about: whether different approaches to "knowledge engineering" actually improve GitHub Copilot's effectiveness with Business Central development. Like many of you working with AI tools, I keep hearing about "feeding knowledge to AI," but I hadn't seen systematic testing of whether it works - and more importantly, which approaches work better than others. So I built a controlled experiment with five different knowledge engineering tiers. The results surprised me. Generic programming knowledge actually _hurt_ performance - we're talking a 4% regression. But when I got the knowledge engineering right? One module went from 64 seconds to 81 milliseconds. That's 746x faster. I've open-sourced the complete testing framework because I'd rather see more people succeed with AI knowledge engineering than keep this approach to myself. Full methodology and results: https://lnkd.in/dhihQAwg I'm curious about your experiences: Have you seen patterns where generic AI guidance actually made things worse? What's worked best in your domain? I've even added instructions about how you could play with it to test different models - tell me what you find!

35

10 Kommentarer

I often think these days about horrible, disfunctional and nightmarish codebases I had to maintain in my past. And then I look at how fast and efficient AI tools generate code at our command. And then I smile. Bitterly. My job is secured. Not sure about yours tho.

11

A challenge to Enterprise Architects I have engaged in a number of conversations lately, especially here on LinkedIn, about software development practices in the age of AI. Yesterday I commented on a post where Dusan Jovanovic said "EA sees the full picture". A link to the post is in the comments. This struck a nerve and I want to share and expand on what I commented in that post. [I have a number of connections and friends who are Enterprise Architects, i.e. they get paid money for being an EA. I'd love to get your feedback on this somewhat ranty post.] I have seen the original of this image from Thomas Nys make its rounds, used to tell various AI stories. Let's focus on this derivative work, "EA sees the full picture". My first thought when seeing this post was, "If EA can see it, why does this picture exist, AI or no AI?" I have worked in enterprise environments and with some great EAs, and I understand the reasons why this picture evolves. But here is me calling out that the Emperor is wearing no clothes. Is not the whole reason for investing in Enterprise Architecture as a function, to define organisational ontology, standards, bounded contexts, etc., and most importantly to lay out invariants that must always hold true for everything deployed? Not only during the design phase, or at first release, but for every single deployment in the lifespan of the thing? I ask with tears in my eyes. I challenge those EAs who are sticking around for this next AI chapter: make yourself relevant if you see this picture. In the same way senior software engineers will need to level up junior engineers to hit the ground running and work at higher levels of abstraction, you need to collaborate closely with the code generators and technology deployers in your business. You need to share this perspective and your knowledge so that all the policy you define is codified into the architectural specification for all software in your environment and other that you introduce. What does provable architectural specifications for software look like? My colleague Martyn Riley has posted an example, the link is also in the comments. We are readying ourselves now for deploying code that is 100% abstracted from the level that software engineers operate at; Martyn calls this SpeC++. Architecture, design, and software "grunt work" are converging. All you modern-day Enterprise Architects, you have a very important part to play.

394

105 Kommentarer

𝐀𝐝𝐨𝐩𝐭𝐢𝐨𝐧 𝐋𝐞𝐬𝐬𝐨𝐧𝐬 𝐟𝐫𝐨𝐦 𝐚 𝐆𝐥𝐨𝐛𝐚𝐥 𝐂𝐨𝐩𝐢𝐥𝐨𝐭 𝐑𝐨𝐥𝐥𝐨𝐮𝐭 !! "Just give people access to AI and they'll naturally maximize its value." This common assumption about M365 Copilot adoption couldn't be further from the truth. Experience from a global Copilot rollout shows that successful adoption follows predictable patterns—none of which happen automatically. The first surprise? Technical aptitude doesn't predict AI success. Many technically skilled users struggled to integrate Copilot effectively, while less technical individuals who were naturally curious and process-oriented quickly became power users. This challenged initial assumptions about who would lead adoption. Clear use cases drive initial engagement. Shifting from generic "explore Copilot" messaging to specific examples relevant to each department—helping sales teams draft proposals, finance teams analyze reports, or HR teams create job descriptions—accelerated adoption dramatically. The most powerful adoption driver wasn't formal training but peer demonstration. Seeing a colleague accomplish in minutes what normally took hours created the emotional response necessary for behavior change. Establishing "Copilot Champions" who demonstrated real-world applications in team meetings amplified this effect. Privacy concerns required direct addressing. Users worried about data security and whether their prompts were being stored or reviewed. Transparent communication about data handling policies alleviated these concerns and removed a significant adoption barrier. AI adoption isn't a one-time event but an ongoing journey. Regular sharing of new techniques, use cases, and best practices maintained momentum long after initial excitement would have naturally waned. What has your experience been with AI tool adoption? Have you encountered resistance or enthusiasm? What approaches have worked in your organization? Share your insights below! #M365Copilot #ChangeManagement #AIAdoption

8

⭐ Book Review: Learn Model Context Protocol with TypeScript by Christoffer Noring “Learn Model Context Protocol with TypeScript” is one of the most practical and forward-looking technical guides available today for developers building AI-powered applications. Christoffer Noring does an exceptional job explaining why MCP (Model Context Protocol) matters, how it evolved, and how it solves real engineering challenges in modern AI systems. The book starts by grounding the reader in the historical background and purpose of MCP, clearly showing why standardized context, structured tool integration, and safe capability execution are essential for today’s LLM-driven applications. From there, it transitions into a deep but accessible breakdown of the MCP architecture, protocol design, hosts, servers, clients, capabilities, and communication layers (STDIO & SSE). The real value shines in the hands-on chapters: Building and testing STDIO servers, including resource and tool development Implementing real-time SSE servers for streaming interactions Streamable HTTP techniques to scale MCP servers for web and cloud consumption Clean architecture patterns for production-ready server design Developing bespoke clients/agents, including LLM-integrated clients Consuming MCP servers through IDEs like Claude Desktop and VS Code The advanced chapters elevate this book beyond a basic tutorial. Topics such as elicitation techniques, sampling strategies, secure communication, authentication/authorization, and production deployment help readers understand how to design robust, scalable, and safe AI workflows — not just prototypes. By the end, the reader gains a complete mental model of MCP and a practical blueprint for building enterprise-strength AI systems that are maintainable, modular, and secure. Some areas could be strengthened with more end-to-end real-world examples and architectural diagrams to tie concepts together. More guidance on common pitfalls, debugging, and error handling would benefit beginners. TypeScript best practices around MCP patterns could be expanded further. And while production topics are covered, deeper deployment examples would make the book even more practical. Verdict: This is a must-read for AI engineers, TypeScript developers, and architects who want to build reliable, extensible, and future-proof agent-based systems. Christoffer delivers both conceptual clarity and hands-on implementation detail — making this one of the strongest resources available for developers working with MCP today. Thanks to Anamika Singh getting this on my radar! Thanks #packt# for this book 😀 #ModelContextProtocol #MCP #TypeScript #AIEngineering #AIDevelopment #LLMTools #AgentDevelopment #StreamingAPIs #Claude #OpenAI #SoftwareArchitecture #AIWorkflowDesign #ServerDevelopment #TechLeadership #AIStandards #ChristofferNoring #BookReview #Developers #LearningJourney #ModernAI #FutureOfDevelopment

11

FlowCentric Processware seamlessly provides insight in average process execution time with exception reporting such as process bottlenecks, or number of reworks per function, process tasks, or user. This enables organisations to optimise processes and eliminate inefficiencies where applicable.

6

Is JSON’s reign over? Why this new format might save your LLM budget. TOON(Token-Oriented Object Notation), a new format designed specifically for feeding structured data to Large Language Models. Check out the spec: https://lnkd.in/g2-Fwa45 ​#LLM #GenAI #DataEngineering #TOON

1