Building Deterministic Infrastructure for Non-Deterministic AI Agents
AI Engineer World's Fair 2026 · 7:14
Social platforms and AI models
Meta operates social platforms including Facebook, Instagram, WhatsApp and Threads, connecting people and selling advertising that helps marketers reach audiences through personalized content. Its Llama model family gives researchers, startups and enterprises a foundation for generative AI applications. Developers can download models, fine-tune them for additional tasks or access them through managed cloud services; Llama 2 on Amazon Bedrock, for example, removes the need to manage underlying infrastructure.
The company began as Facebook in 2004, founded by Mark Zuckerberg, Dustin Moskovitz, Chris Hughes, Andrew McCollum and Eduardo Saverin. Zuckerberg remains chairman and CEO, and the parent company adopted the Meta name in 2021. Llama originated within Meta’s FAIR research organization, where researchers explored how relatively small language models trained with curated data and appropriate scaling laws could support new applications. Subsequent community work extended instruction following, language coverage and context windows.
Meta went public in 2012. In September 2023, the company reported more than 30 million downloads of Llama-based models through Hugging Face and over 3,500 enterprise project starts across Google Cloud and AWS. Its 2026 teen safety agreement covering Facebook and Instagram includes approximately $18 billion in payments over ten years, with 30% conditional on specified actions by YouTube and TikTok.
AI Engineer World's Fair 2026 · 7:14
AI Engineer Europe 2026 · 18:00
Affiliations reflect their AIE appearances, not necessarily current employment.
Start here to understand how retry amplification and elastic workload scheduling affect the cost and operation of long-running agents.
Nishant GuptaAI Engineer World's Fair 2026
Continue here to learn how distributed tracing and continuous evaluation help assess agents during production operation.
Nishant GuptaAI Engineer World's Fair 2026
Choose this demonstration to see how lowering creation barriers still leaves designers with decisions about what makes a game compelling.
Danielle An · David HoeAI Engineer Europe 2026
Gupta treats autonomous agents as distributed systems, examining how stateful workflows, shared-memory consistency, and operational controls can address unpredictable model behavior and runaway resource use.
Gupta argues that offline accuracy misses production behavior. His evaluation discussion covers tool use, failure recovery, and scenario-driven task completion through reliability metrics and production telemetry.
Danielle An and David Hoe describe using key art to anchor an LLM, testing ideas through rapid prototypes, and building multiplayer experiences with runtime-LLM-driven NPCs.
Affiliations reflect each recorded session, not necessarily current employment.