Technical requirementsThe gap between hype and realityThis distinctionLLM applications vs agentic systemsThe characteristics of LLM applicationsThe characteristics of agentic systemsThe operational implicationsPrompt chainingThe agent loop: Reason, Act, ObserveReasonActObserveWhy this loop is crucialTool use and function callingModels and toolsFunction callingTool DesignsMCP and reusable tool interfacesThe joy of context engineeringThe context is your moatAgent lifecyclesThe five-phase lifecycleSeparation of concernsThe human implications of this separationShort-lived and long-lived agentsHello Google ADKWhy ADK?ADK core primitivesA simple ADK agent: Design choices, not syntaxWhat ADK does not doGoogle AI stack terminologyThe layeringGemini models as reasoning enginesModel optionsThinking levelsKey capabilitiesModel selectionConsumption options: How you pay matters as much as what you runGemini in production: Governance constraintsThe operational challenge aheadThe governance gapAgentOps, EvalOps, and the Agent Development LifecycleCommon anti-patterns"Agent in name only""Prototype is production""The omniscient agent"AgentOps pressure testDecision playbook: Is this an AI-powered or agentic system?Summary