An archive of decisions, failures, changes, and shipped work. The record stays useful by keeping the sequence—and the limits—visible.
A LiveKit hotel voice-agent demonstration raises a practical question: when a caller changes the dates after a quote, what exactly are they agreeing to?
The hotel voice agent had a plausible booking ready to go: October 10–12, two guests, a king room with an ocean view. It quoted $582.40 and asked whether the caller wanted to proceed.
Claude Code mods make the runtime question concrete: who can alter an agent's prompt, tools, permissions or display after an extension has been admitted?
A state-topology lens for deciding what parallel agents should inherit, share, or isolate, and for exposing the consistency, provenance, authority, contention, and recovery costs of each choice.
An analysis of why agent-to-agent communication needs explicit identity, delegated scope, task lifecycle, cancellation, failure, and evidence semantics once a message can trigger consequential work.
An analysis of when another agent worker adds enough distinct value to justify the context, contention, verification, and reduction costs it creates.
A runtime change can preserve an agent's name, memory, and files while leaving a harder question unresolved: which history, authority, state, and behavior actually carried forward?
A pre-fan-out test for deciding when agent specialization or parallelism is worth the context a handoff can discard or force the system to reconstruct.