‘We are looking at capturing sentiments with AI agents’: Google’s Bikram Singh Bedi on what’s next after chatbots

Google along with developers are now moving on developing “AI agents” that can take actions independently without human intervention, marking a step toward automation. Simply put, these are advanced generative AI tools capable of performing multistep, complex tasks on a user’s behalf and generating their own to-do lists, going beyond simply creating responses like chatbots. If these agents, as envisioned by companies like Google, can perform tasks automatically and are deployed in the right places, they could fundamentally change how we work.

Bedi tells indianexpress.com AI agents are already being deployed by companies in India. “When I look at some of the work we at Google are doing with our customers in India, I believe we have moved beyond the pilot stage,” he says. Bedi cites the example of Federal Bank’s ‘FEDDY’, an AI-powered virtual assistant. Ever since the bank introduced ‘FEDDY’, customer care costs have decreased by 50 per cent, and NPS has increased by 25 per cent, with virtual queries doubling.

“An AI agent is addressing internal productivity challenges as well as customer care issues. The implementation may vary by vertical; for example, the approach could be different for an insurance company compared to a bank, even though both are in financial services. Now, consider the impact in areas like HR, or in large companies such as Google, or a major Indian manufacturing company.

Google announced the integration of its AI model Gemini with Beckn-enabled open networks and apps at the 2024 Google for India event in New Delhi on October 3, 2024. (Image: Anuj Bhatia/The Indian Express)

Google, Microsoft, and OpenAI are all preparing AI agents designed to be fully customisable and automate more complex, multi-step tasks for both enterprise operations and consumer functions. Earlier this year, at Google Cloud Next, the company introduced a new tool to help businesses build AI agents. Its Vertex AI Agent Builder, a no-code solution, combines Vertex AI Search and Conversation products, expanding the developer’s toolkit for building generative AI agents.

“The app will essentially embed that agent, which will connect to all the various networks at the back end. It’s an open-source platform, not just Google working on it, but the community will also be involved in its development.”

But Bedi continues to press on Google’s unique advantages. “Google is the only vendor out there which does all of this end-to-end. Every other vendor is either buying GPUs, marking it up and selling it, or outsourcing other AI technology, marking it up and selling it. Some don’t have first-party services. We built this whole stack,” he continues.

There are, of course, safety concerns when it comes to AI agents, in addition to the risks already inherent in generative AI like hallucinations, bias, the risk of fabrications as they interact with each other, and latency due to the complex processes required from agents. Even though Big Tech companies and startups are pouring resources into developing AI agent systems, widespread deployment is still likely some time away, warn experts.

“We will not be using any data from any company, or network, without explicit consent,” Bedi clarifies when asked if Google trains the data on what the company is plugging into the database.