Meta was kind enough to extend an invitation for me to attend their inaugural LlamaCon—a one-day developer summit devoted to the Llama family of open-source large language models. It offered the chance to better understand the direction in which both the technology and its surrounding ecosystem are moving, and therefore merits a close read by anyone shaping AI strategy or policy.
Open Source Gaining Momentum
Mark Zuckerberg highlighted the rapid expansion of the open-source AI community since Llama’s launch, with downloads surpassing 1.2 billion, up from 650 million last December. He noted that a year ago, there were only a few open-source options, but now there are a growing number of open models from Google, Mistral, DeepSeek, and soon, OpenAI. Meta’s Chris Cox noted that thousands of developers are actively creating tens of thousands of derivative models. For example, Nvidia’s new Llama-3.1 Nemotron Ultra outperforms DeepSeek’s R1 despite being half its size. Meta also announced a Llama AP, which will help make it easier for organizations to deploy AI capabilities. Databricks CEO Ali Ghodsi also highlighted how Crisis Text Line uses a customized version of Llama to identify individuals at risk of self-harm or suicide. The platform, having engaged in millions of crisis conversations, leverages Llama to enhance the precision and effectiveness of its risk assessments.
Voice: The Interface After Touch
Zuckerberg highlighted voice as the next major interface for AI, noting that ultra-low latency will be key to delivering natural, real-time interactions, especially in wearable tech like the Meta Ray-Ban smart glasses.
I completely agree. Voice capabilities are currently underestimated despite how remarkably human-like AI voices have become. Engaging AI systems in this way is more like Tony Stark talking to Jarvis. It provides a more intuitive and natural experience and unlocks exciting possibilities in education, customer service, healthcare, and other fields.
The Rise of AI Agents
AI agents came up in nearly every session. Zuckerberg and Microsoft CEO Satya Nadella noted that about 30 percent of their organizations’ code is already AI-generated. Zuckerberg expects most project code will soon be written entirely by AI, producing higher-quality outputs at a faster rate than human developers.
That matters because Washington’s policy discourse is still fixated on chatbots. In contrast, nearly every conversation I have in Silicon Valley now revolves around AI agents that can reason, plan, take action, and reflect with a high degree of autonomy.
This trend toward agent-centric AI shifts the use away from just query-response interactions toward active collaboration with intelligent digital coworkers—yet its implications are still largely not discussed in DC. The leap from chatbots to agents is not incremental—it’s a paradigm shift, and policy that misses this risks both under-protecting society and over-constraining innovation.
Redefining Digital Content
Nadella raised a deceptively simple question: When one interface can generate text, code, images, and runnable simulations—what is a “document”?. ChatGPT, Google Gemini, Meta.ai, and Anthropic Claude each offer a “canvas” capable of generating diverse content, ranging from basic text to complex images and functioning code. Today, you can chat with a PDF to extract key insights, launch a deep research dive across multiple sources, and then use that same interface to generate an interactive simulation—all within a single AI-powered environment. I don’t know what you call that, but it’s an emerging form of content that could shake up traditional publishing, especially in education.
Strategic Reflections
- Geopolitical Stakes of Open Source AI: The release of DeepSeek R1 in January underscored the growing strategic importance of open-source frontier AI, not just as a technological advancement, but as a key factor in US–China competition and American national security. I would much rather countries and organizations in the global south use US-based open source models than embed Chinese models into their systems and infrastructure.
- Expertise-as-a-Service: Humans augmented by AI defined the last two years of GenAI. Now, we are at the early stages of AI agents emerging as true digital co-workers. Open models will democratize access to expertise and intelligence, reaching millions globally. This shift isn’t merely software-as-a-service: It’s “expertise-as-a-service.” Microsoft’s recent report highlights this critical transition, and policymakers must grapple seriously with its implications.
- Policy & Civil Society Engagement: Meta deserves credit for including public policy and civil society representatives at the event, which helps bridge the gap between technology and policy, a practice more AI companies should adopt to ensure responsible and informed policymaking.
Meta’s first-ever LlamaCon highlighted the remarkable growth and strategic importance of open-source AI, underscoring its potential to democratize access to intelligence. Open models are emerging as critical assets, enabling broader access to advanced AI capabilities and influencing geopolitical competition and policy decisions worldwide. The organizations—and governments—that internalize these shifts fastest will shape the rules, reap the talent, and set the pace for the decade ahead.
The post Agents, Access, and Advantage: Lessons from Meta’s LlamaCon appeared first on American Enterprise Institute – AEI.
