Leif Larsen
Software architect & tech lead | AI, architecture and the pace of work | ex-Microsoft
Oslo, Norway
Actions
Leif Larsen is a software architect and tech lead at Novanet, working mostly in .NET and Azure. He has spent a good part of his career untangling systems that grew one reasonable decision at a time. In 2025 he was at Microsoft, helping development teams adopt GitHub Copilot, and these days he builds most of his side projects together with AI. That's where his opinions on AI agents come from, the enthusiastic ones and the worried ones.
He also wrote a book on Microsoft's AI services, mostly at five in the morning before work, which is how he ended up with opinions on burnout, and on what AI is doing to the pace of work.
When he's not at a keyboard, he's running, dabbling with music, or losing an argument with pizza dough.
Area of Expertise
Topics
From 100 applications to 1 platform
Nobody sets out to build a hundred applications. You get there one reasonable decision at a time. "Yes, we can do that!" "We'll use our CRM for that." A few years later you are sitting on a large on-premise estate, with a CRM doing jobs it was never meant to do and integrations running in every direction that nobody fully understands.
That is where a large Norwegian organization was when we started. This talk is the story of how we turned it into one platform on Azure, while the business kept running as usual. We put abstractions in front of the old core systems so they could be replaced without the rest of the organization noticing, built the new services on .NET and Azure one piece at a time, and shut the old applications down as we went. That last part sounds obvious. It is also the easiest one to skip.
It was not all smooth. A renewal like this is expensive and hard to estimate, and someone has to pay for it. We had to slow down before we could speed up, and we spent quite a bit of time deciding which legacy bugs to fix and which ones to live with until the system was gone. I'll be honest about what worked and what didn't.
You will also see the architecture we ended up with and the stack behind it, from API Management and Service Bus to .NET minimal APIs, a library of Bicep modules and integration tests with Testcontainers.
For architects, tech leads and .NET developers sitting on a legacy estate of their own, or trying not to build one.
Your AI Agent Is a Liability. Prove Me Wrong.
Everyone is building AI agents. Fewer people are asking whether they should. In this session I make the case that autonomous agents are a default architectural mistake: they increase fragility, destroy debuggability, and make failures harder to reason about than the deterministic systems they replace.
I am not anti-agent. I am anti-cargo-cult.
Drawing on a concrete decision framework, I will show you exactly where agents earn their complexity and where they do not. We will work through the operational line between a chatbot and a true agent, three diagnostic questions every team should answer before building, and real examples of where agents fail catastrophically in production.
You will leave with a strong default bias toward simpler architectures, and a clear, defensible test for when breaking that default is actually justified.
NDC AI 2026 Sessionize Event Upcoming
HelloStavanger 2026 Sessionize Event
TDC 2024 Sessionize Event
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top