Serious AI happens in boring .NET, not in clean demos
Where the system can't be switched off, auditability stops being optional.
There's a curious asymmetry in the market: almost all the public discussion about applied AI happens on modern stacks, and almost all the work that's worth a quote happens on stacks nobody wants to show at a conference.
The constraint is the advantage
A .NET system that has been running for twelve years has characteristics that all look like drawbacks: it's rigid, badly documented, and nobody can switch it off. But it also has something new projects don't: a stream of real, repetitive, expensive decisions, with an owner who knows exactly what they cost.
That changes the conversation. You don't start from "what could we automate", a question that never converges, but from "we make this decision four thousand times a year and it costs us this much".
The enterprise constraint isn't an obstacle to the project: it's its specification.
And auditability comes free in the conversation
In the enterprise world nobody needs convincing that traceability is required: it's already a requirement for other reasons. The decision ledger isn't a feature to sell, it's a box someone was already trying to tick.