Let's continue our trek through agentic devsecops practices. Last week we discussed the agentic playground and the access control it brought as a substrate for leveraging partially trusted users: LLMs. Today we'll dive into the nuts and bolts.

First, let's bound the solution space properly with some requirements, the final setup must be:

But what about prompts? Agentic development has an enormously protean character in that it blends flow of thought and conversation with engineering. For that I developed a dedicated ontology.

Agentic sessions are forests we walk through without being able to walk back.

next, one way to walk all that talk