Live demo
Inside an AI agent, one step at a time
tokens in
tokens out
tokens in
tokens out
AGENT HARNESS
Customer
Sara, web chat
Identity & auth
shared API key
Guardrails
rules, filters, policy
Agent loop
think → act → observe
Context engine
AGENTS.md
Skill: refunds
Memory
RAG: policies
Model
tokens in → tokens out
MCP: Orders
orders.get
MCP: Refunds
refunds.create
Logistics agent
remote, via A2A
Human approver
support lead
Traces & evals
every step recorded and graded
PRESENTER NOTES