Why it matters
Guardrail and tool-restriction layers are exactly what stands between an agent and runaway API spend, so this is useful even though it's framed as a safety topic.
The tokenmaxxing angle
Permission controls and tool restrictions are the mechanisms that cap what actions, and therefore how many model calls, an agent can take, so guardrail design is effectively agent-cost design.
From the organizers
Page (retitled "AI Guardrails in Education") says each session is a paper review or technical deep dive on system prompts, policy enforcement and tool restrictions, at The Frontier RTP, hosted by Daniel S.