AI that speaks, codes and takes action: Inside Google Cloud’s Gemini masterclass

At Google Cloud's TechSparks 2026 event, Rajesh Shivani, Head of AI Customer Engineering, showcased how Google's Gemini AI system is evolving beyond answering questions to taking on complex tasks. Shivani demonstrated how Gemini can process 80,000 files to create presentations with charts, images and narratives, or build detailed 3D scenes using code without special rendering software.

He explained how AI is moving from simple chatbot interactions to agentic AI, where systems can plan tasks, access information, and act on user behalf.

Agentic AI requires four key elements: a goal, clear instructions, relevant knowledge and tools to execute. Access to a company's internal data is crucial for building useful agents, as general-purpose models may lack access to specific processes or customer information. Google's Open Knowledge Format (OKF) aims to provide businesses with a structured way to make their information accessible to AI agents.

Voice AI was highlighted as a key demonstration area, with systems developed for Singapore-based Grab helping passengers and drivers coordinate airport pickups despite language barriers. The demo showed voice AI's potential to support interactions in over 85 languages. Another example featured AI dubbing an English video into Marathi and Malayalam while synchronizing lip movements and facial expressions.

These capabilities could streamline the production of multilingual training material, marketing videos and customer communications.

Shivani also showcased Google's tools for coding and cybersecurity. Antigravity assists developers in building applications, while CodeMender identifies code vulnerabilities, generates patches and tests them in a sandbox. In one demonstration, Gemini analyzed 80,000 files from NASA's open archives, creating a presentation using a supplied template and design guidelines. Another showed a model generating a detailed 3D library scene using Python and NumPy, without the need for dedicated rendering software or external assets.

For businesses considering AI adoption, Shivani emphasized that powerful models are not always necessary. Simpler tasks can be handled by faster, lower-cost models. As AI agents take on more work, businesses will need to rethink permissions, security, costs and automated activity. Shivani urged organizations to prepare for a future where agents, not just people, interact with their systems, and developers become managers of agents rather than handling every step themselves.