Google Agent Development Kit, usually called ADK, is a code-first framework for creating, evaluating and deploying agents. It supports model-driven agents alongside deterministic workflow agents and can connect external tools through OpenAPI and MCP.
Key takeaways
- Supports model agents plus sequential, parallel and loop workflow agents.
- Designed around tools, sessions, artifacts, evaluation and deployment.
- Model and deployment options extend beyond one Google service, although Google integrations are prominent.
What is Google Agent Development Kit?
Code-first agent development kit from Google for tool-using agents, workflow agents, multi-agent systems and deployment across supported runtimes. ADK fits teams that want explicit workflow primitives and a clear path from local development to Google-oriented deployment without making every step model-directed.
What can you build with Google Agent Development Kit?
- Build LLM agents and deterministic workflow agents.
- Compose multiple agents with delegation and shared session state.
- Use custom functions, OpenAPI tools, MCP tools and supported Google services.
- Evaluate locally and deploy to supported managed or self-hosted targets.
These are documented capabilities, not a guarantee that every model, provider or deployment supports the same behavior. Validate the exact SDK version, model features and tool permissions in a disposable environment before moving a workflow into production.
What is a sensible first project?
Build a sequential research workflow that accepts one approved URL, extracts a structured brief and routes it to a review agent. Disable arbitrary web navigation and store the evaluation trace without credentials or full sensitive documents.
Keep the first run narrow and observable: one input, a small tool allowlist, explicit success criteria, a cost ceiling and a human review point before any external write. Save the prompt, model, SDK version, tool arguments and final result so the test can be reproduced.
How does the architecture handle state and tools?
Agents run inside an ADK runtime with sessions, events and artifacts. Workflow agents control order and repetition deterministically, while LLM agents choose tools and delegate according to instructions. Runners and services provide the execution boundary.
Treat model output as untrusted input. Validate structured data, set timeouts and iteration limits, make write operations idempotent where possible, and separate read-only discovery from actions that modify files, infrastructure, customer records or messages.
What should you review before deployment?
- Enable tool confirmation for operations with side effects.
- Scope cloud service accounts to the exact resources used by the agent.
- Separate session and artifact retention policies from chat history.
Use least-privileged credentials and isolate code execution, browsers and shell tools. Log tool calls without recording secrets, define an emergency stop, and test how the application behaves when the model, a tool or the network returns an error. Human approval should be enforced in application code for high-impact actions rather than requested only in a prompt.
What are the main limitations?
- Feature parity may differ across ADK language implementations.
- Google deployment examples can hide provider-specific operational choices.
- Loop and delegation limits still need application-level enforcement.
This profile is based on public first-party documentation checked on 2026-10-04; Anavem did not run a comparative benchmark or a production deployment. APIs, package names, licensing boundaries and hosted services can change, so confirm the current documentation before adopting the framework.
Is Google Agent Development Kit the right choice?
Choose it when its programming language, orchestration model and operational controls match a concrete workflow. Compare it with one simpler baseline, including a direct model API plus ordinary application code. The useful decision is not which framework has the longest feature list, but which one makes tool permissions, state, failure handling, evaluation and maintenance understandable to your team.