Agent Anchor
Keeping production AI agents from quietly getting worse: a loop that notices when quality drops, finds the cause, and checks the fix.
CMU capstone · BNYagentsllmcode private
What it is
An industry capstone with BNY. AI agents in production can quietly get worse as the models, prompts and tools underneath them change, and today a person usually notices before the system does. Agent Anchor is a loop that catches it first.
The loop
- Notice a real drop in quality.
- Diagnose what changed.
- Fix it.
- Verify the fix didn’t break anything else.
My part
I work on both ends: deciding what should count as an agent getting worse, and checking that a fix actually helps before it ships.