Articles &
field guides

Trace Brief treats AI agents as systems rather than product labels. The archive follows the parts that determine whether an agent works in production: model and runtime choices, retrieval, memory, tool protocols, identity, evaluation, security controls, and recovery from failure. Each field guide starts with a concrete engineering question, separates demonstrated capability from vendor claims, and links technical conclusions to primary documentation or research.

Use the architecture guides when you are defining boundaries between models, tools, state, and human approval. Protocol explainers are for teams deciding how agents should discover capabilities or communicate across systems. The security and operations articles focus on permissions, traceability, evidence, and failure handling. If you are comparing two approaches, begin with the constraint that matters most—latency, reliability, control, or interoperability—then follow the related links inside each article instead of choosing a stack from feature counts alone.