I reproduce agents failures and shrink them to a minimal case. So the main question: Where do I find the teams that need this?

I'll describe what I actually do, then ask for advice, because I'm stuck on the customer part. I run a small one-person service for teams that build AI agents. You know the failure mode. An agent works in demos, then a customer sends a transcript where it did something crazy, and nobody can reproduce it. Or the eval harness tells you a regression happened somewhere between Tuesday and Thursday, across two hundred prompts, and that's all you get. Here's what I do with that. First I reproduce the failure under strict conditions: the case set and the call budget are frozen before the first model call, and every case is scored by two independent oracles that have to agree. Adapter errors stay separate from wrong answers, so a timeout never gets counted as the model being dumb. Then I shrink the failing case with a budgeted search: the minimal case still fails, every smaller candidate passes. Then you fix it, and I verify the fix against the minimal case and hand you a test you can run yourself. An example from my own records, not client work. A cloud model failing an 8-operation state-machine case shrank to 2 operations in 33 calls. The original and the 2-op case both fail 3 out of 3 runs. Every 1-op candidate passes. That's the deliverable: not "the model has issues", but the smallest input that fails, plus a test that catches it. The method works. The business side doesn't yet. I've sent emails to a dozen B2B integrators and agent teams. Zero replies. So maybe you could advise where I should focus my attention and search for test customers?