AI coding agents write code fast, but they also drift: they bend rules, skip tests and blur architectural boundaries. This session shows how to build .NET Core applications where the architecture itself keeps agents (and humans) honest and trustworthy.

I will show you a recipe of a pure business core in F# and everything around it in familiar C#: ASP.NET Core minimal APIs, Marten for event sourcing and Wolverine for CQRS.
The core uses the Decider pattern (decide and evolve), and business failures are always valid values, not exceptions.
A thin facade exposes it to C#, which works well with F# unions.

Nothing of this is science fiction or vanishing words.
I've coded a real after-sales and repair application, for my company; we will look at the guardrails I've setup to make this work in an agentic workflow:

Executable specifications: Given / When / Then scenarios and property-based tests in F#, no mocks and no I/O.
Integration testing with self managed containers.
Architecture tests that fail the build when the domain depends on infrastructure or becomes mutable.
And of course even more for the harness.

You will leave with a structure you can get inspired, an honest view of the costs (tooling, onboarding, interop friction) and a rule of thumb for when this approach is worth it.