Build practical Google ADK agents that can survive real implementation constraints.
A rigorous, hands-on professional workshop focused on implementing a bounded Google Agent Development Kit system rather than surveying generic agent theory. Learners use a version-pinned official ADK reference implementation—normally Python 2.x for the shared lab—while recording language-specific assumptions and checking documented parity before transferring patterns to Go, Java, JavaScript, or Kotlin. Across six modules, participants define an agent boundary, connect a meaningful capability, implement justified orchestration, engineer state and human controls, evaluate and observe execution, and produce a deployment-minded implementation and handoff package. Mastery requires a minimum score of 80% across implementation correctness, architectural rationale, safety and control, evaluation evidence, operability, and handoff quality.
Workshop structure: 6 modules · 18 applied topics · 6 Decision Labs · 7 assessments · 80% mastery · approximately 6–8 hours.
Applied outcome: build and defend a working ADK architecture with tools, orchestration, sessions/state, evaluation, observability, failure handling, and a deployment-minded handoff.
- Implement a bounded Google ADK agent system with explicit instructions, model choice, runner behavior, events, sessions, state, deterministic application logic, and documented interfaces.
- Create testable ADK tool contracts for custom functions and a verified integration path such as OpenAPI or MCP, with explicit credentials, permissions, errors, side effects, and human approval boundaries.
- Select and implement the simplest justified orchestration pattern, including single-agent delegation, sub-agent composition, deterministic workflows, routing, parallel execution, loops, or A2A only where requirements warrant them.
- Engineer session, state, memory, artifact, callback, approval, recovery, identity, secret, and permission controls without assuming persistence, durability, or security properties that have not been configured.
- Build representative tests and evaluation datasets, review trajectories and traces, instrument supported telemetry, and define measurable quality, latency, cost, release, rollback, and maintenance criteria.
- Deliver and defend a deployment-minded ADK implementation and handoff package that achieves at least 80% on the workshop mastery rubric.
Your strongest mastered Decision Lab work is collected in your personal Google ADK Agent Build Portfolio.