Latin America's First National-Scale AI Agent

The Argentine government officially launched 'MIA,' a public service AI agent based on Meta's open-source large language model (LLM) Llama 4, on September 23. MIA represents the evolution of the existing chatbot system TINA into generative AI, built upon infrastructure that handles an average of 1.5 million monthly citizen consultations and integrates with 72 government agencies.

At the launch event held at the Cultural Science Center, Minister of Innovation, Science and Technology Darío Genua stated, "Introducing AI to public administration means more than improving service delivery—it signifies moving toward citizen-centered, rapid, and innovative governance." The project involved collaboration between the government, Meta, and Argentine chatbot platform company Botmaker.

From TINA to MIA: What's Changed

The Argentine government's official chatbot TINA has served as the digital touchpoint for public services since its 2020 launch. Currently, TINA supports over 600 administrative procedures, handling an average of 290,000 monthly consultations with human agents and facilitating 200,000 document downloads.

MIA's core differentiator is contextual understanding and personalized responses through generative AI. While TINA responded according to predetermined scenarios, MIA can understand user intent, respond in natural language, and guide users through complex administrative procedures step-by-step.

Initial Services: Startup/Export Support and Mi Argentina Integration

In Phase 1, MIA focuses on supporting SMEs and startups. The AI directly provides information needed for business activities, such as business registration procedures, export regulations, and government support program guidance. This strategy aims to reduce time lost due to administrative complexity in the early startup stages.

Integration with the Mi Argentina app (26 million users) is also planned. Currently, the app's help desk handles over 80,000 monthly inquiries, and MIA is expected to significantly reduce agent workload by handling initial responses. The government stated this could "save citizens thousands of hours annually spent waiting for administrative procedures."

Technology Choice: Why Llama 4?

The Argentine government chose Meta's Llama 4 for three main reasons.

1. Cost Efficiency
According to government explanations, Llama 4's operating costs are one-tenth those of commercial models like GPT-4 or Claude. Lower cloud infrastructure and inference costs make it suitable for large-scale public services.

2. Open-Source Flexibility
As an open-source model, Llama 4 allows the government to operate the model on its own servers and fine-tune it as needed. This avoids data sovereignty issues and external service outage risks that could arise from commercial API dependency.

3. Scalability and Performance
Llama 4 is known for multilingual support and fast response times, and the Argentine government determined it could be operated stably even when scaled nationwide.

Botmaker co-founder Alejandro Zuzenberg evaluated MIA as "capable of understanding and responding in natural language on infrastructure that prioritizes privacy protection, audit trails, and local data management," describing it as "a responsible innovation model replicable throughout Latin America."

Global Context of Public Sector AI Adoption

Government-level generative AI adoption has spread globally since 2023. Estonia introduced the 'Bürokratt' AI assistant to public services in late 2023, and Singapore operates 'Pair,' a generative AI tool for civil servants. The U.S. federal government has been running AI pilot programs across departments following a 2024 Biden executive order.

Argentina's MIA is differentiated by being open-source model-based public AI. While most government AI projects depend on commercial APIs from OpenAI, Anthropic, or Google, Argentina operates Llama on its own infrastructure, securing data control. This could attract attention as an AI strategy model for financially constrained countries.

However, challenges of bias, errors, and accountability in public AI adoption remain. If generative AI provides incorrect information, citizens could be harmed, particularly in sensitive areas like law, taxation, and welfare. Argentina's initial focus on the relatively less sensitive area of startup/export consulting can be interpreted as risk management.

[AI Analysis] Next Steps for Public AI: Expansion and Challenges

MIA's success depends on expansion speed and accuracy management going forward. If initial startup consulting shows positive results, the scope could expand to welfare applications, tax inquiries, healthcare service guidance, and more. Particularly once Mi Argentina app integration is complete, AI services will be directly exposed to 26 million users, formally shaping citizen perceptions of public AI.

Technically, Llama 4's open-source nature could be a double-edged sword. While the government can reduce costs by operating the model independently, it must also manage model updates, security patches, and performance improvements internally—tasks that commercial API providers handle automatically.

In a broader context, Argentina's case offers implications for AI sovereignty strategies in developing countries. It demonstrates that open-source ecosystems can be leveraged to build AI services for citizens without depending on major AI companies from the U.S. or China. However, for long-term sustainability, this requires parallel investments in AI talent development, data infrastructure, and legal/ethical frameworks.

Public AI has shifted from a question of "whether to adopt" to "how to operate responsibly." MIA will be one case study in that experiment.