Google ADK Assess

agentsopen-sourceplatformgooglevertexassessaimay-2026

Overview

Google Agent Development Kit (ADK) is an open-source framework that structures agents with tools, memory, and deployment paths on Google Cloud, where Vertex AI has been brought into the Gemini Enterprise Agent Platform (ADK docs, ADK on Gemini Enterprise Agent Platform, Google Cloud Next '26).

Assess when your estate is GCP-first. Validate portability of agent definitions if multi-cloud is a requirement.

Adoption Signals

  • ADK 2.0 reached general availability for Python on 19 May 2026, Go on 30 June 2026 and TypeScript on 21 August 2026, introducing graph-based workflows (ADK 2.0, ADK Python v2.0.0 release).
  • Google announced the Gemini Enterprise Agent Platform at Cloud Next on 22 April 2026, bringing Vertex AI together with Agent Studio, Agent Registry, Agent Identity and Agent Gateway, and named the ADK's new graph-based framework as part of it (Google Cloud Next '26).
  • The Python package had about 9.9 million PyPI downloads in the 30 days to 30 September 2026 (PyPI Stats), and the Python repository has about 21,700 GitHub stars (adk-python on GitHub).
  • ADK includes an optimization framework: adk optimize applies the GEPA algorithm to refine root agent instructions against an evaluation set (ADK agent optimization).

Risks

  • Breaking 2.0 migration. ADK 2.0 adds node_info and output fields to the event schema, which breaks custom session stores with rigid columns and strict JSON validators, and makes BaseAgent a workflow node so that custom overrides of 1.x execution methods are silently ignored (ADK 2.0 compatibility notes).
  • Google Cloud gravity. The managed runtime, registry, identity and gateway features sit in the Gemini Enterprise Agent Platform (Google Cloud Next '26); agents built around them are harder to move to another cloud even though the framework itself is open source.
  • Uneven language parity. Python leads; the ADK 2.0 page lists 2.0 for Python, TypeScript and Go, while the Java SDK is still on the 1.x line (ADK 2.0, adk-java on GitHub).
  • Release pace. adk-python went from 2.0.0 in May to 2.10.0 in September 2026 (adk-python releases), so pinned versions age quickly.

Pros & Cons

Advantages

  • Open source under Apache-2.0 with generally available 2.0 releases for Python (May 2026), Go (June 2026) and TypeScript (August 2026), plus a Java SDK, so polyglot teams can share one agent model.
  • ADK 2.0 adds a graph-based workflow runtime for deterministic, conditional and looping agent flows, alongside built-in evaluation, A2A support and an adk optimize command that applies GEPA to agent instructions.
  • It is the code-first framework of Google Cloud's Gemini Enterprise Agent Platform, with direct deployment paths to Agent Runtime, Cloud Run and GKE.

Disadvantages

  • The 1.x to 2.0 upgrade changed the event schema and the agent execution contract; custom overrides of 1.x methods such as _run_async_impl are silently bypassed by the new workflow engine.
  • Gemini and Google Cloud get the deepest integration; other models are supported through adapters, and some built-in tools and deployment features only make sense on Google Cloud.
  • Release cadence is fast (Python reached 2.10 four months after 2.0), which means frequent upgrades and re-testing for teams that pin versions.

Recommendation

Google ADK is rated Assess for most teams. It is worth a structured evaluation where Google Cloud is the primary platform and Gemini the primary model family, because that is where the framework, the managed Agent Runtime and the governance features of the Gemini Enterprise Agent Platform fit together. Build one representative agent on ADK 2.0 directly, not on 1.x, and compare it with the framework you already use on evaluation results, tracing quality and effort to deploy.

For multi-cloud or model-agnostic requirements, test a non-Gemini model through ADK's model adapters early and keep business logic in plain functions and MCP or A2A services rather than in ADK-specific classes. Pin versions, run the ADK evaluation suite in CI, and read the release notes before each upgrade, since the 2.0 migration showed that execution-contract changes can fail silently.

Sources

Consulting

From radar to implementation

You have assessed the technologies — we help you bring them into production systems safely and measurably.

AI agents

AI agents

Agents in production, not just in a prototype: RUBINLAKE AI Agents handle research, forecasting and market monitoring for your team. Source-backed and operated by us.

Learn more
AI Trainings

AI Trainings

We make your teams AI-ready. Our trainings turn abstract buzzwords into applicable skills. Clear. Practical. Built for doers.

Learn more