What candies are Agentic AI frameworks stealing from developers?

The rise of Agentic AI frameworks marks a profound change in software engineering, fundamentally shifting the boundary between what developers write and what a framework handles. But what, exactly, are these powerful frameworks automating away?
This provocative talk, "What candies are Agentic AI frameworks stealing from developers?", explores how modern Agentic frameworks abstract key parts of the system flow. We will clarify and reproduce core agent responsibilities, such as tool orchestration and the Agent control loop, demonstrating exactly how these are being absorbed by the framework. This is not just a high-level discussion; it is a practical lesson on framework design and architectural knowledge essential before building agents. By analyzing which "candies" are being "stolen," attendees will gain the clarity needed to focus their future skill development in where it's being expected that developers invest their craft.

Workshop - Build Multi Agent Systems with A2A

In this hands on session, you will go beyond basic chatbots to build a distributed multi agent system powered by the Agent to Agent (A2A) open source protocol.

You will build a Course Creation Agent Pipeline using the Agent Development Kit (ADK), an open source and code first framework from Google Cloud designed to simplify the development, evaluation, and deployment of complex AI agents. Each agent maintains a single responsibility, collaborating toward a shared objective and self correcting until the goal is achieved.

This workshop provides a practical foundation for designing and implementing robust multi agent architectures.

By the end of this workshop, you will:

- Master the fundamentals of ADK and multi agent systems.
- Design agents with specialized, clear responsibilities.
- Test and run agents locally.
- Deploy a multi agent application to Google Cloud.
- Launch your first publicly available multi agent system.

Requirements:
- Level: Beginner friendly; no prior multi agent experience required.
- Skills: Basic experience with Python, Google Cloud, Docker, and Bash is a plus.
- Credits: Provide a PERSONAL Gmail account at least three days prior to receive free Google Cloud credits.
- Setup: Bring a charged laptop and avoid using a VPN during workshop tasks.

Deployed app: https://www.carneiro.dev/ws-multiagent

Build a distributed multi agent system using the open source Agent to Agent (A2A) protocol and the Agent Development Kit (ADK). This beginner friendly workshop guides you through designing specialized agents, implementing self correcting feedback loops, and deploying a live Course Creation Pipeline. Participants will move from local testing to a public cloud deployment.

! Please provide a personal Gmail account three days in advance for receiving Google Cloud credits and bring a charged laptop without a VPN !

Talk - Building Interoperable Agents: When A2A meets MCP

It is time to move from isolated chatbots to ecosystems of collaborating agents. But how do we ensure agents can talk to each other without building proprietary silos?

In this session, we will break down interoperable AI using two powerful open standards recently donated to the Linux Foundation: the Agent-to-Agent (A2A) protocol and the Model Context Protocol (MCP).

We will start with a live demo of a deployed multi agent system, where an orchestrator agent coordinates the collaboration of multiple agents using A2A, while some of those agents share an MCP server. After the demo, we will deconstruct the system architecture. While explaining these Apache 2.0 licensed protocols, we will walk through the engineering process to rebuild the presented agentic system from scratch.

Join to see how A2A and MCP work hand in hand to create a new, open platform for Agentic AI, and why you should jump right in!

Deployed app: https://www.carneiro.dev/ducktaipe
GitHub: https://github.com/carneiroDotDev/ducktAIpe

It is time to move from isolated chatbots to ecosystems of collaborating agents. But how do we ensure agents can talk to each other without building proprietary silos?

Join to see how A2A and MCP work hand in hand to create a new, open platform for Agentic AI, and why you should jump right in!