Modern artificial intelligence systems are capable of performing complex tasks that go beyond simply answering queries. An example is provided where AI processes 80,000 files and generates a presentation with charts, images, and clear narration, or creates a detailed 3D scene using code without specialized rendering software.
At the TechSparks 2026 event, Rajesh Shivani, Head of AI Client Engineering for Digital Natives, Startups, and IT Companies at Google Cloud, presented how Gemini is applied in the fields of voice AI, autonomous agents, and software development.
His masterclass focused on how enterprises can transition from using AI to receive answers to delegating work tasks to it. Initially, AI tools gained popularity through chat interfaces that answered questions. The next stage is agentic AI, which allows systems to plan tasks, access information, and use tools to act on behalf of the user.
Rajesh explained that 'an agent is now just a model plus a framework,' describing the infrastructure around the model that manages memory, context, permissions, and task execution. Creating useful agents requires four components: goals, clear instructions, relevant knowledge, and tools to apply that knowledge. Internal company data is particularly important because general-purpose models may not have access to this company's specific processes or customer information.
Shivani also discussed Google's Open Knowledge Format (OKF), which aims to provide organizations with a structured way to make information accessible to AI agents.
One key demonstration example was the use of voice AI. A system developed for the Singaporean company Grab helped passengers and drivers coordinate meetings at the airport, overcoming language barriers. This demonstration showed how voice AI can support interaction in over 85 languages.
Another example demonstrated AI's ability to dub videos in English into Marathi and Malayalam while simultaneously synchronizing lip movements and facial expressions. Such features can help companies in producing multilingual educational materials, marketing videos, and customer communications with fewer production steps.
Rajesh Shivani emphasized that 'voice is becoming the new user interface,' pointing to the growing importance of voice interaction for customers and employees.
He also presented a virtual banking assistant created for Citibank's retail banking. During the demonstration, the AI avatar explained options for repaying a time deposit and processed an extension after the customer selected a 12-month term.
In addition to voice functions, Shivani highlighted Google's tools for coding and cybersecurity. The Antigravity tool helps developers build applications, and CodeMender can detect vulnerabilities in code, generate fixes, and test them in a sandbox.
In one demonstration, Gemini analyzed 80,000 files from open NASA archives, including Mars weather data and helicopter telemetry, and prepared a presentation using a provided template and design guidelines.
Another demonstration showed how the model generated a detailed 3D library scene using Python and NumPy, without the need for specialized rendering software or external resources.
For companies considering the costs of AI implementation, Shivani noted that the most powerful model is not always necessary. He explained that 'you don't need a cutting-edge model for everything,' as simpler tasks can be handled by faster and less expensive models.
As agents take on more work, enterprises will need to rethink their methods for managing permissions, security, costs, and automated activities. Shivani strongly urged organizations to consider a future where their systems interact with agents, not just humans. He suggested that developers will increasingly be able to act as agent managers instead of performing every step themselves.
The main challenge is no longer just selecting a capable AI model, but providing agents with the necessary knowledge and tools while ensuring they perform tasks reliably and securely.
