@vukosi We’ve sped up how fast code gets written, but not how fast features ship. Review and change management remain human dependent, and that’s where the delay now sits.
@MABUwaAI I’m well aware of the evals framework, both Opik built in and custom. I’m more concerned around building a framework for “red/blue teaming” a framework to test implemented guardrails and producing the evidence where they pass/fail.
@MABUwaAI Have you considered a framework for Agent AI use cases that takes your guardrails spec as input, derives adversarial tests per control - retrieval injection, tool misuse, prompt injection - and fails/flags the build when one breaks. Similar to pen tests for traditional software.