The Agent Platform and ADK
Explore what Gemini Enterprise Agent Platform provides across building, scaling, governing, and optimizing agents. Learn the core building blocks of the Agent Development Kit, including agents, tools, workflow agents, sessions, and callbacks, and how to enforce business rules in code.
We'll cover the following...
When a customer moves from a single Gemini API call to a production agent, the engineering problem changes. The system now needs to manage conversations across many turns, call tools safely, keep memory between sessions, run reliably at scale, and give security and operations teams the controls they expect. Building all of this from scratch for every engagement would slow delivery and introduce risk. Google’s answer is a managed platform for the surrounding infrastructure and an open-source framework for the agent logic itself.
Gemini Enterprise Agent Platform provides the managed infrastructure, and the Agent Development Kit (ADK) provides the framework. FDE interviewers expect candidates to know what each one is responsible for, where the boundary between them sits, and how an ADK agent is structured. The sections below cover the platform’s capabilities first and then the core building blocks of ADK.
What the Agent Platform provides
Google introduced Gemini Enterprise Agent Platform in April 2026 as the evolution of Vertex AI. It brings model access, agent building, runtime, governance, and optimization into one platform. Google groups its capabilities into four areas, and this grouping is a useful way to explain the platform in an interview.
Area | Purpose | Key capabilities |
Build | Create agents and choose models | Agent Development Kit, Agent Studio (low-code), Agent Garden (templates), Model Garden |
Scale | Run agents reliably in production | Agent Runtime, Agent Sessions, Agent Memory Bank, Agent Sandbox |
Govern | Control identity, access, and risk | Agent Identity, Agent Registry, Agent Gateway with Model Armor, threat and anomaly detection |
Optimize | Measure and improve agent quality | Agent Simulation, Agent Evaluation, Agent Observability, Agent Optimizer |
Each area answers a question that enterprise customers ask during a deployment.
Build answers how the agent is created and which model it uses.
Scale answers where the agent runs and how it remembers context. Agent Runtime hosts agents as a managed service, Agent Sessions stores the history of a conversation, and Agent Memory Bank keeps useful information across separate conversations.
Govern answers who the agent acts as and what it is allowed to do. Agent ...