AI agents at scale: Why devs need to build with Google Cloud

AI agents are moving from simple copilots to autonomous teammates capable of reasoning, planning, and collaboration. This was the key theme at the recently concluded Hyderabad edition of DevSparks 2025, YourStory’s flagship event for the developer ecosystem, which aims to empower one million Indian developers with the tools and insights to build global-scale products.

Madhu Shekar, Customer CTO at Google Cloud, brought these ideas to life in his Lightning Tech Talk on ‘Building agentic systems with Google Cloud’. In his presentation, he introduced the Agent Development Kit (ADK), the Agent2Agent (A2A) protocol, and Vertex AI Agent Builder, tools designed to help developers build, manage, and scale AI agents for real-world use.

From ChatGPT to Gemini

Shekar began his session by reflecting on where we are in the AI journey. While the launch of ChatGPT in 2022 brought generative AI into the spotlight, Google had already been advancing the field with BERT, PaLM, and now Gemini. With billions of parameters and multimodal capabilities spanning text, voice, and video, Gemini represents a leap towards richer, more collaborative interactions.

“A significant part of software engineering is scaffolding, error handling, and test scripts, the mundane work that can easily be automated,” Shekar explained. Generative AI is already acting as a pair programmer, but the next leap is asynchronous agents that can take on tasks independently, essentially becoming another engineer on the team.

The maturity of agents

Shekar outlined four stages in the maturity journey of AI agents: automation of routine tasks, reasoning and planning, continuous learning, and finally, multi-agent collaboration. He pointed to Gemini Astra as an example of what is possible today. On a Google Pixel phone, Astra guided a bicycle repair by pulling up the right YouTube video, identifying the required screw, and overlaying instructions, showing how multimodal agents can combine reasoning, vision, and context in real time.

Jules in action

If Astra demonstrates the future, Jules is a glimpse of what developers can use right now. Shekar described Jules as “an AI software engineer” capable of plugging into GitHub, detecting bugs, and making the code changes.

“It is a completely independent, asynchronous software agent,” he said. “You give it a task, and it comes back with a solution ready to merge with your repository.” By operating in the background, Jules allows human developers to concentrate on design and strategy while repetitive work is handled automatically.

Building blocks for agentic systems

Shekar highlighted how Google Cloud’s ecosystem supports this shift. The ADK, now part of the Linux Foundation, lets developers build and extend agents with custom workflows. The A2A protocol enables agents to communicate and collaborate, while the Vertex AI Agent Builder provides a secure managed environment for building, testing, and deploying agents at scale.

“These frameworks are what prevent chaos when companies start building hundreds of agents across functions,” he warned. “With ADK and Agent Engine, you get the guardrails needed for governance, cost control, and security.”

Why this matters for developers

Shekar urged developers to take prompt management seriously. Saving, versioning, and reusing prompts, he said, is now as important as source control in coding. “If you don’t save your prompts, you are losing context,” he noted.

For developers, the rise of agents is an opportunity to shape the next generation of enterprise workflows, whether in IT, marketing, or customer service, while ensuring discipline around reproducibility, security, and quality.

He closed with a call to action, encouraging developers to experiment with Gemini Live and NotebookLM, and to join the GenAI Exchange Hackathon, where over 45,000 developers are already building. With mentorship from Google experts and a prize money of Rs 65 lakh up for grabs, the hackathon is open to students, startups, enterprises, and freelancers.

For developers, the opportunity is clear: to embrace frameworks like ADK, A2A, and Vertex AI Agent Builder, and to start building agents that are secure, production-ready, and capable of working alongside humans as true digital teammates.

“The sooner you get hands-on, the sooner you’ll see how agentic AI can transform the way you build and solve problems,” concluded Shekar.