Reliable agent-driven development

Two Topics on todays agenda:

Reliable agent-driven development
(Thomas Martinsen, Tech Evangelist @ Twoday, MS MVP & RD)
Agent-driven development is not about letting tools write code for you. It is about shifting responsibility. We move from writing every line ourselves to defining intent, constraints, architecture, and security boundaries—while agents produce the first drafts.

The real challenge is not generation, but reliability and value. Tools will evolve. What matters is understanding how they work, where they fail, and how to design guardrails around them. Just as anyone can write code but not everyone practices solid software engineering, the difference in this new paradigm lies in discipline, not access to AI.

This session focuses on how to make agent-driven development robust: how to define clear requirements, enforce non-functional constraints, embed security and observability, and validate AI-generated artifacts at scale. The goal is to raise the abstraction level without losing control—and to ensure that agent-generated output translates into sustained business value.

How Does a Coding Agent Work?
(Tore Nestenius, .NET Architect & Technical Trainer, MS MVP)

Coding agents are increasingly used to automate software tasks, yet how they actually work is often treated as a black box. This talk breaks a coding agent down into its core mechanics: how it is structured, how it communicates with a language model, and how a text-only model can read and modify real source code.
We will cover:
*what a coding agent is, mechanically
* how agents communicate with language models
*how tools and MCP calls are executed
* how a text-only model works with source code
* how to build a basic coding agent in C#