# Season 2 — March 2026 (Spring Cherry Blossom Edition)

Canonical Web URL: https://adventofagents.com/2026/03/

- **Day 01: [Season 2 Kick Off](https://adventofagents.com/2026/03/01?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day01)** ([.md](https://adventofagents.com/2026/03/01.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day01)) — 31 days. Zero to Production-Ready AI Agents on Google Cloud.
- **Day 02: [Build ADK Agents with Gemini 3.1 Pro](https://adventofagents.com/2026/03/02?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day02)** ([.md](https://adventofagents.com/2026/03/02.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day02)) — Choose your language. One command. Total language flexibility. Bootstrap high-performance agents in seconds.
- **Day 03: [Build AI Agents with Gemini 3.1 Flash-Lite](https://adventofagents.com/2026/03/03?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day03)** ([.md](https://adventofagents.com/2026/03/03.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day03)) — Announcing Gemini 3.1 Flash-Lite: our most cost-efficient model for building high-quality AI agents at scale.
- **Day 04: [MCP Servers: Add external tools to your agents in just a few lines](https://adventofagents.com/2026/03/04?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day04)** ([.md](https://adventofagents.com/2026/03/04.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day04)) — Wire up multiple MCP servers simultaneously to trace bugs from Linear to GitHub PRs.
- **Day 05: [Long Term Recall: Memory Plugins](https://adventofagents.com/2026/03/05?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day05)** ([.md](https://adventofagents.com/2026/03/05.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day05)) — Implement persistent semantic memory for AI agents using the GoodmemPlugin.
- **Day 06: [ADK Skills](https://adventofagents.com/2026/03/06?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day06)** ([.md](https://adventofagents.com/2026/03/06.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day06)) — Eliminate wasted tokens with progressive disclosure. Use ADK Skills to load complex instructions only when needed.
- **Day 07: [ADK Agent Skill Design Patterns](https://adventofagents.com/2026/03/07?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day07)** ([.md](https://adventofagents.com/2026/03/07.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day07)) — Five design patterns for structuring SKILL.md content in ADK — Tool Wrapper, Generator, Reviewer, Inversion, and Pipeline.
- **Day 08: [Multi-Agent Patterns: Sequential Agents](https://adventofagents.com/2026/03/08?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day08)** ([.md](https://adventofagents.com/2026/03/08.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day08)) — Build predictable sequential workflows where the output of one agent goes straight to the next.
- **Day 09: [Multi-Agent Patterns: Coordinator/Dispatcher Agents](https://adventofagents.com/2026/03/09?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day09)** ([.md](https://adventofagents.com/2026/03/09.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day09)) — Build high-fidelity educational video agents that maintain character and scene consistency using the Google ADK, Gemini 3.1 Pro, Nano Banana and Veo 3.1.
- **Day 10: [Multi-Agent Patterns: Parallel Fanout and State Interpolation](https://adventofagents.com/2026/03/10?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day10)** ([.md](https://adventofagents.com/2026/03/10.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day10)) — Drastically reduce latency by running independent, grounded LLM tasks concurrently and automatically synthesizing their outputs.
- **Day 11: [Multi-Agent Patterns: Hierarchical Decomposition](https://adventofagents.com/2026/03/11?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day11)** ([.md](https://adventofagents.com/2026/03/11.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day11)) — Use a top-level Manager agent that utilizes an AgentTool to dynamically generate a plan before executing it.
- **Day 12: [Multi-Agent Patterns: Generator-Critic Agent Loop](https://adventofagents.com/2026/03/12?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day12)** ([.md](https://adventofagents.com/2026/03/12.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day12)) — Automatically refine LLM outputs using ADK's LoopAgent to orchestrate a conversation between a creative Writer and a strict Critic.
- **Day 13: [Multi-Agent Patterns: Iterative Refinement](https://adventofagents.com/2026/03/13?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day13)** ([.md](https://adventofagents.com/2026/03/13.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day13)) — Compose Skills, MCP, and Code Execution into a meta-agent that builds, tests, and refines other ADK agents through iterative loops.
- **Day 14: [Multi-Agent Patterns: Human in the Loop](https://adventofagents.com/2026/03/14?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day14)** ([.md](https://adventofagents.com/2026/03/14.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day14)) — Inject an approval breakpoint to pause agent execution and wait for manual user confirmation before executing sensitive financial APIs.
- **Day 15: [Grounding with ADK: Agentic RAG with Vector Search 2.0](https://adventofagents.com/2026/03/15?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day15)** ([.md](https://adventofagents.com/2026/03/15.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day15)) — Build an Agentic RAG system in 3 minutes with Vector Search 2.0 and ADK.
- **Day 16: [ADK Dev Skills: Accelerated Multiagent Triage](https://adventofagents.com/2026/03/16?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day16)** ([.md](https://adventofagents.com/2026/03/16.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day16)) — Accelerate multiagent system development from scaffolding to deployment using ADK Dev Skills.
- **Day 17: [Workspace & Gemini Enterprise: no-code agents](https://adventofagents.com/2026/03/17?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day17)** ([.md](https://adventofagents.com/2026/03/17.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day17)) — Connect and use Google Workspace connectors in personal Gemini Enterprise agents.
- **Day 18: [Workspace & Gemini Enterprise: ADK agents](https://adventofagents.com/2026/03/18?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day18)** ([.md](https://adventofagents.com/2026/03/18.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day18)) — Build ADK agents that use the Vertex AI Search MCP server and Google Workspace APIs, deploy them to Vertex AI, and register them with Gemini Enterprise.
- **Day 19: [Live Shopping Agent: Build with ADK and Gemini Embedding 2](https://adventofagents.com/2026/03/19?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day19)** ([.md](https://adventofagents.com/2026/03/19.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day19)) — Build a live multimodal shopping app with ADK and Gemini Embedding 2.
- **Day 20: [ADK Agent Harness: Build a Generate-Validate-Refine Pipeline](https://adventofagents.com/2026/03/20?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day20)** ([.md](https://adventofagents.com/2026/03/20.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day20)) — Build a README improvement harness that fetches a GitHub repo via MCP, generates against a quality checklist, and refines in a loop until a critic approves.
- **Day 21: [Developer's Guide to AI Agent Protocols](https://adventofagents.com/2026/03/21?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day21)** ([.md](https://adventofagents.com/2026/03/21.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day21)) — Six protocols, disambiguated. See how MCP, A2A, UCP, AP2, A2UI, and AG-UI work together to build a production-ready supply chain agent with ADK.
- **Day 22: [ADK Evaluation: Trajectory Tests and Rubric-Based Scoring](https://adventofagents.com/2026/03/22?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day22)** ([.md](https://adventofagents.com/2026/03/22.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day22)) — Define deterministic trajectory tests and rubric-based evaluations that run on every code push, catching agent regressions before they reach production.
- **Day 23: [Model Armor: AI Security Firewall for Agents](https://adventofagents.com/2026/03/23?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day23)** ([.md](https://adventofagents.com/2026/03/23.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day23)) — Protect AI agents from prompt injection, jailbreaks, and data leakage using Google Cloud Model Armor as a defense-in-depth security layer.
- **Day 24: [Batch Processing: Scale to 10k with ADK](https://adventofagents.com/2026/03/24?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day24)** ([.md](https://adventofagents.com/2026/03/24.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day24)) — Shift from interactive processing to an Agent as Orchestrator pattern using the ADK and Gemini Batch API for efficient, large-scale asynchronous workloads.
- **Day 25: [Agent Deployment: How to Deploy AI Agents](https://adventofagents.com/2026/03/25?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day25)** ([.md](https://adventofagents.com/2026/03/25.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day25)) — Deploy your AI agents to Vertex AI Agent Engine or Google Cloud Run securely and easily.
- **Day 26: [Authentication: End-User Identity Propagation](https://adventofagents.com/2026/03/26?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day26)** ([.md](https://adventofagents.com/2026/03/26.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day26)) — Securely delegate permissions by prompting end-users for OAuth consent during agent execution.
- **Day 27: [Scion: an open testbed for agent orchestration](https://adventofagents.com/2026/03/27?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day27)** ([.md](https://adventofagents.com/2026/03/27.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day27)) — Explore multi-agent patterns with an agnostic supra-harness system that isolates agents in git worktrees and containers, allowing easy orchestration and communication
- **Day 28: [A2A Protocol: Decoupling Reasoning from Execution](https://adventofagents.com/2026/03/28?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day28)** ([.md](https://adventofagents.com/2026/03/28.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day28)) — Decouple your reasoning engine from execution using the universal A2A 1.0 protocol to bridge a Python LangGraph orchestrator and a Go ADK service.
- **Day 29: [ApiRegistry: Dynamically Fetching BigQuery Tools](https://adventofagents.com/2026/03/29?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day29)** ([.md](https://adventofagents.com/2026/03/29.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day29)) — Learn how to use the ApiRegistry object to dynamically fetch an admin-approved, fully configured BigQuery tool directly from the Cloud API Registry at runtime.
- **Day 30: [Observability: Debug with Hierarchical Tracing](https://adventofagents.com/2026/03/30?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day30)** ([.md](https://adventofagents.com/2026/03/30.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day30)) — Eliminate silent failures in AI agents by implementing Hierarchical Trace Observability with OpenTelemetry and Arize Phoenix.
- **Day 31: [A2UI & A2A: Building the Dog Weather App](https://adventofagents.com/2026/03/31?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day31)** ([.md](https://adventofagents.com/2026/03/31.md?utm_source=adventofagents&utm_medium=markdown&utm_campaign=adventofagents_s2_2026&utm_content=day31)) — Transform static agent responses into rich interactive micro-apps natively rendered in Gemini Enterprise using A2UI.