Overview
This lecture introduced the Agentic Revolution, focusing on why now is the ideal time to build intelligent AI agents, the evolution of agent technology, and the tools provided by Google Cloud to empower startups and developers in creating agentic systems.
Program & Session Structure
- The Startup School offers a multi-week curriculum, with live and recorded sessions, hands-on labs, and practical resources.
- Week 1 covers why it's time to build intelligent agents; later weeks focus on the agent engine and hands-on agent development.
- Labs and badges are available for practical experience via Cloud Skills Boost.
Introduction to AI Agents
- Early chatbots had narrow, single-purpose use with limited context and memory.
- Modern agents (conventional and agentic) are autonomous, goal-oriented, and may work collectively to solve complex problems.
- Agentic systems handle multiple modalities, tasks, and can orchestrate multiple agents/tools.
Why Agents, Why Now?
- Most businesses have a SaaS (software) layer and a human (operational) layer; the human layer is labor-intensive and inefficient.
- AI agents can augment human work, especially within vertical-specific tasks, presenting a major opportunity for startups.
- Recent evolution in AI: LLMs (Large Language Models) → Retrieval-Augmented Generation (RAG) → Tool integration → Multi-agent systems.
Technological Advancements Enabling Agents
- Constant improvements in AI models: more frequent releases, reduced usage costs, and multimodal input/output support.
- Enhanced tooling: frameworks like LangChain, LangGraph, Crew AI, and Google's own ADK.
- Robust platforms: end-to-end solutions for building, testing, and deploying agents (e.g., Google Vertex AI).
Google Cloud Agentic Tools
- Gemini: Family of multimodal foundation models (text, image, audio, video inputs/outputs), with Pro and Flash variants.
- Agent Development Kit (ADK): Modular framework for easy agent development and orchestration, integrates with multiple LLMs and tools.
- Vertex AI: Comprehensive machine learning platform for data prep, model building, fine-tuning, evaluation, and scalable agent deployment.
- Agent Engine: Manages scaling, security, session state, logging, and deployment of agents, compatible with various frameworks and models.
Key Demos & Features
- Building a travel assistant agent in minutes using Gemini in AI Studio.
- Gemini 2.5 supports native audio input/output and emotional voice responses, not reliant on traditional text-to-speech.
- ADK simplifies agent creation by combining Python functions and models into deployable agents.
- Vertex AI Model Garden offers access to 200+ models, both proprietary and open source.
Q&A Highlights
- Non-engineers can build agents via no-code or low-code platforms like Firebase Studio.
- Deployment of agents is simplified via integrated publishing features (e.g., Firebase Studio, Agent Engine).
- Gemini generates native audio responses, now with humanlike emotional tone, across many languages.
Key Terms & Definitions
- AI Agent — Autonomous, goal-driven software that acts without ongoing human intervention.
- Agentic System — A collection of agents working collaboratively using tools and multimodal data.
- LLM (Large Language Model) — AI model trained on vast text data, used for content generation and understanding.
- RAG (Retrieval-Augmented Generation) — Technique combining LLMs with data retrievers for updated, external info.
- ADK (Agent Development Kit) — Google’s toolkit for building and orchestrating AI agents.
- Vertex AI — Google’s platform for end-to-end machine learning and agent deployment.
- Agent Engine — Service for managing, scaling, and monitoring deployed AI agents.
Action Items / Next Steps
- Review week zero session if missed.
- Try AI Studio (https://a.dev) and experiment with building simple agents.
- Explore resources in the AI Agent Handbook (link in chat/QR code).
- Apply for Google Cloud credits if you’re an eligible startup.
- Prepare for the next session: “Architecting Intelligent Agents” on June 12th.