Presenters
Source
AI at Scale 2026: The Agentic Revolution is Here! 🚀
The world of AI and data is evolving at lightning speed, and the AI and Data at Scale 2026 conference, hosted by Meta’s Engineering and Infrastructure teams, was a testament to this rapid transformation. From planet-scale systems to the dawn of agentic autonomy, this event offered a profound glimpse into the future of technology.
The Compute Craze Unleashed by AI 🤯
The exponential pace of change in AI has unleashed an unprecedented appetite for compute. The industry is building data centers at a scale never seen before, and this demand shows no signs of slowing down. Meta itself has invested $115 million in the America’s Workforce Academy to bridge the growing talent gap. We’re seeing record numbers of code generated by AI, and agentic token consumption is skyrocketing, growing orders of magnitude quarter-over-quarter.
The Core Product Loop: Evolving with Agents 🔄
The traditional product development loop, from consumer interaction to recommendation systems powered by large language models, is being fundamentally reshaped by agents. Before agents, all these functions were meticulously designed, built, and operated by human talent. Now, agents can code, manage data, analyze reports, and even build other models. The next frontier? Expanding agent autonomy, allowing them to perform tasks for hours, days, or even weeks, with humans stepping in to guide and supervise. This paradigm shift moves humans from in the loop to over the loop.
Agents as Infrastructure Citizens: A Paradigm Shift 🌐
The industry is witnessing an inflection point: automated traffic has surpassed human traffic on the internet (51% and growing eight times faster!). Inside our own infrastructure, a similar inversion is happening. Agents are no longer just features; they are first-class consumers of infrastructure, changing everything.
The Challenges of Agentic Infrastructure 🚧
For two decades, infrastructure was optimized for human request patterns—predictable, session-based, with natural rate limits. Agents, however, don’t sleep, eat, or get distracted. This leads to several breaking points:
- Capacity Breaks: One engineer now spawns tens, even hundreds, of agents, leading to a load increase equivalent to 100,000 employees overnight for a thousand-person organization.
- Identity Breaks: Agents lack human identities, badges, or deployment records, making them distinct from services or batch jobs. They make decisions, posing new identity management challenges.
- Velocity Breaks: While GitHub Copilot writes 46% of code, the CI/CD pipelines haven’t sped up accordingly. Agents write code in seconds, but the subsequent build, test, deploy, and monitor stages remain a bottleneck.
The core question for today’s conference is: Can your infrastructure handle what happens when agents do the work?
Data: The Heart of the Agentic Revolution ❤️
Agents are making a significant impact on data. In just three months, 63% of dashboards published at Meta were created by agentic data apps, fueling a 30x growth in agent queries. Democratizing data access is a win, but it raises crucial questions about trust and governance.
The Trust Paradox in Data Access 🤔
Traditionally, human experts curated and analyzed data, understanding privacy and regulatory aspects. When agents take over, especially with less context, how do we trust the data? Do they understand regulations? Can we backtrack if something goes wrong? This is a major risk, even as humans remain involved.
The Data Deluge: Training Data Demands 📈
The amount of data required is growing exponentially. The ratio of training data to active parameters in open-weight LLMs has grown 3.1 times per year since 2022. Recent models use 20 times more data per parameter than recommended by Chinchilla scaling laws. This evolution demands a rethinking of data infrastructure:
- Freshness Matters: Real-time streaming data, not batch ETL, is becoming the backbone for ranking and recommendation pipelines.
- Intelligent Storage: Storage layers must understand the data they serve, fetching only necessary columns and time ranges in flexible formats.
Recommendations: From Pattern Matching to Reasoning 💡
Recommendation systems, the heart of many businesses, are evolving from pattern matching to genuine reasoning. LLMs are breaking the paradigm, allowing systems to understand user intent, not just clicks.
- The Northstar: Conversational Recommendations: Imagine algorithms you can direct, telling them what you want more or less of, and they reason about your intent in real-time. This leads to fully conversational recommendations that understand requests, not just keywords.
The Flywheel of Innovation: Agents, Data, and Reasoning 🎡
The entire product loop—consumer, product, data, and recommendations—is being transformed. Agents are becoming citizens of the loop, operating, discovering, analyzing, and creating alongside humans. Data is shifting from batch to real-time, from opaque blobs to intelligent storage. Recommendations are evolving from pattern matching to reasoning. This creates a powerful flywheel where each advancement accelerates the others.
Emerging Agentic Innovations at Meta 🚀
Meta is exploring a range of agentic applications:
- Task Agents: Working on coding, data analysis, and operations.
- Clawtown: A multi-agent orchestration layer enabling semi-autonomous decisions for weeks.
- Moldbook: An agent-to-agent social network where humans observe agent communication and negotiation.
- Frontier Loop: Accelerating executive decision-making by autonomously synthesizing context from critical documents.
The Next Frontier: Serving Billions of People and Millions of Agents 🌌
The Atscale conference started a decade ago discussing systems for a billion people. Today, the conversation has shifted to serving billions of people and millions of super-intelligent agents simultaneously. When models train themselves and continuously adapt, and when agents and humans co-create at this scale, we won’t see 10% improvements; we’ll witness entirely new categories of products. The talks at this event are not academic exercises; they are a glimpse into the future being built today.
The question remains: What will scaled infrastructure enable you to build?
This comprehensive blog post synthesizes the key insights from the AI and Data at Scale 2026 conference, highlighting the transformative power of agentic AI, the challenges and opportunities in infrastructure, data, and governance, and the exciting future that lies ahead.