Localiza Reorganized Work Around Generative AI, With 60+ Agents Behind Liza

Hi everyone, it's Shii-chan!

OpenAI News openai.com

Here's a Localiza story from OpenAI News. It's about changing how people work, not just adding a tool.

What was announced?

OpenAI News published a story about how Brazilian mobility company Localiza reorganized work around generative AI.

Localiza treats generative AI as a change in how the business operates, not as another software tool. The collaboration with OpenAI rests on three pillars:

• Experimentation at scale internally
• New customer interfaces
• AI-native products

Why it matters

CTO André Petenussi describes AI as a tool that helps teams do what they already do well, and says the main obstacle is human, not technical: teaching a large organization new ways of working.

The technology operation alone has 19 tribes and 130 squads, so spreading experiments at that scale is notable.

What changes

1. Scale enables learning

• ChatGPT licenses were distributed as an experimentation and productivity resource, not an individual perk
• Teams in areas like finance and daily operations run experiments with custom GPTs, agents, and Codex
• Codex started in technology and spread to analysts, managers, PMOs, and lawyers, who build their own prototypes and automations
• An unexpected effect: multidisciplinary teams now prototype before work enters the development queue

2. A new customer entry point

• Customers increasingly state needs in natural language, and traffic from LLMs has grown rapidly since 2024, with ChatGPT as the main channel
• In March 2026, Localiza became the first mobility platform in Brazil to launch an app integrated directly into ChatGPT (one of the first in Latin America). Users explore rental, subscription, and used-car options in the conversation, then complete the transaction in Localiza's secure environment
• Localiza later joined ChatGPT's advertising pilot in Brazil as an advertiser

Dive Deep

The third pillar is Liza, a conversational assistant.

• Unlike a traditional chatbot with a fixed decision tree, it is built around conversation, intent, and context
• It combines OpenAI models, Localiza services, and specialized tools, from the initial request through the rental journey
• More than 60 specialized agents built on reasoning models orchestrate the journey with MCP (Model Context Protocol) integration and specialized tools
• Voice-to-text and text-to-voice capabilities extend the interface
• Success is judged mainly by customer experience quality, with conversion as a secondary metric
• Because Localiza had mapped and measured its service journeys first, it could decide where an agent acts and where a human steps in
• Aggregated signals from interactions reveal needs and friction points that feed Liza's evolution

The collaboration began early: the first version of Liza ran on GPT-3.5.

Tips from the article:

• Start from the customer problem
• Treat licenses as an experimentation budget
• Set quality metrics before scaling agents
• Use each interaction's signals to improve the product

Next up are the next version of Liza, sales-focused agents, conversational interfaces as a customer entry point, and agentic software development.

Wrap-up

• Licenses across ChatGPT Enterprise, the OpenAI API, and Codex serve as an experimentation resource
• A ChatGPT-integrated app launched in March 2026 opens a new customer entry point
• Liza runs on 60+ specialized agents with MCP integration
• The biggest hurdle is people and organizational learning, not technology

This is for anyone who wants AI to change how a whole company works, not just one team!