Google launches Gemini 3.8 Live: AI agents are moving from chat interfaces to real-time digital workers

Google recently introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. On the surface, this looks like another upgrade to real-time voice AI. But from a product and startup perspective, the more important shift is the emergence of a new way for AI agents to work: they can keep talking with users in real time while continuing to reason, call tools, and move tasks forward in the background.

Gemini 3.8 Live focuses on low-latency, natural voice interaction, making it suitable for customer support, voice assistants, language learning, and other scenarios where fast responses matter. Extended Thinking goes a step further by combining real-time conversation with more complex reasoning, allowing the model to continue multi-step analysis and asynchronous tool use without interrupting the flow of the conversation.

Until now, voice AI has often had to choose between speed and depth. Simple questions could be answered quickly, but more complex tasks often required noticeable waiting while the model reasoned or used external tools. This new approach separates the two processes: the front end keeps the conversation natural, while the back end continues thinking and executing.

That changes the product form of Voice AI.

Most AI assistants today still follow a simple pattern: the user asks, and the AI responds. The next generation of agents will increasingly understand the user’s goal, break the task into steps, query systems, use tools, take actions, and continue the conversation throughout the process.

For startups, the opportunity is not simply to build another voice chatbot. The larger opportunity is to turn voice into a real business execution layer.

Imagine a sales agent speaking with a customer while simultaneously checking CRM records, reviewing previous conversations, identifying the customer stage, checking inventory and pricing, generating a quote, requesting internal approval if needed, updating the CRM, sending an email, and scheduling the next follow-up.

From the customer’s perspective, it may feel like a normal conversation. Inside the company, however, an entire business workflow may already have been completed.

This is why AI Plus Lab sees an important transition taking place: Voice Agents are moving from customer-service tools toward digital workers. Earlier generations of voice AI mainly focused on answering questions. The next generation will increasingly be judged by whether they can actually complete work.

This shift could also change how people interact with software.

Today, using enterprise software usually means opening an application, finding the right menu, filling out forms, and clicking through multiple steps. In the future, a user may simply say:

“Find out why this customer still hasn’t renewed.”

The agent could then review CRM data, order history, support tickets, and emails. The user might continue:

“Create a new renewal proposal with a 5% price increase compared with last year.”

The agent generates the proposal, sends it through approval, and completes the next steps.

The software still exists, but more of its functionality may become hidden behind the agent.

At the same time, core model capabilities are becoming infrastructure. Speech recognition, voice generation, real-time understanding, and multi-step reasoning are increasingly being packaged by companies such as Google, OpenAI, and Anthropic.

For startups, this means the competitive advantage will gradually move away from the model itself and toward industry knowledge, workflow design, enterprise data, tool integrations, permission systems, approval mechanisms, memory, security, auditability, and reliable last-mile execution.

In other words, the most important question for an AI agent product may no longer be:

Which model does it use?

It may become:

What work can it actually complete for the user?

From this perspective, Gemini 3.8 Live is not just a better voice experience.

It is another signal that the next generation of digital workers may not live inside a chat box. They may become real-time agents that users can talk to, meet with, delegate work to, and rely on to keep moving tasks forward in the background.

The end state of Voice AI may not be a chatbot that sounds more human. It may be a digital worker that can communicate in real time, keep thinking, and actually get work done.

— AI Plus Lab