Who hurts
Teams with an agent in production and no idea whether the last prompt edit helped.
The problem
Prompt changes are shipped on vibes. A tweak that fixes one complaint silently breaks six behaviours nobody has a test for, and it surfaces as a support ticket a week later.
What you build
Capture production traces, let the team promote any trace to a test case with an assertion in plain English, then replay the whole suite on every prompt or model change and diff the outcomes — pass, fail, and 'changed but still fine'.
Why now
Every team that shipped an agent in the last two years is now on their third model migration and has no regression story. The pain arrives on a schedule.
Validate it this week
Offer to run one migration by hand for a team on a model deadline. Charge for the report. If they pay for the report, they'll pay for the tool.
Why you'd keep winning
The captured trace corpus belongs to the customer but lives in your schema. Once a suite has 300 promoted cases, leaving means rebuilding it.
The honest risk
LangSmith and Braintrust are already here with funding. Win on the migration workflow specifically, not on being another observability dashboard.
The prompt is written to make an AI argue with you before it writes code — that first round of pushback is worth more than the scaffold.