Traceability is now an interface for AI

When building multi-agent AI systems, developers always hit a wall with context management. Giving AI too much data causes it to hallucinate and creates total chaos when agents run in parallel. It also burns through tokens, making development way too expensive.

This talk covers our current R&D work. We are adapting a strict, proprietary systems analysis framework: Grigory Tsiperman’s Method of Adaptive Clustering (MAC).

We are testing a simple hypothesis: requirements traceability doesn't have to be passive documentation anymore. Instead, it can act as a solid interface to set strict boundaries for AI context.

In this session, we will share our conceptual model for turning MAC's decomposition levels into prompt contracts for AI orchestrators. We will also show how we designed our MVP experiment.