Study Track
1.Study the decisions, not the vocabulary
60 min2.Choose the smallest surface that carries the workload
75 min3.Spend capability where errors are costly
75 min4.Turn a request into a testable contract
75 min5.Place each fact in the right type of context
75 min6.Validate claims, not confidence
75 min7.The Messages API is a state machine
90 min8.Structured output is an untrusted contract by default
75 min9.A tool loop is controlled delegation
90 min10.MCP decouples capability from host
75 min11.The Agent SDK is a harness, not a permission slip
90 min12.Security lives outside the prompt
90 min13.Evals turn agent behavior into engineering evidence
75 min14.Claude Code scales through shared constraints
75 min15.Multi-agent orchestration is about when NOT to use multiple agents
90 min16.Resume or fork without carrying stale context
75 min17.A tool error is also a response
75 min18.Scope is part of the answer
75 min19.A single good result is not evidence of reliability
75 min20.Make long context observable
75 min21.Defend an architecture across six contexts
120 min