Rethinking Legacy Data Infrastructure with Eon Co-Founders Ofir Ehrlich and Gonen Stein

The Critical Role of Data in the AI Era

Eon co-founders Ofir Ehrlich and Gonen Stein, in conversation with Elad Gil, emphasize that data is the most valuable asset an organization possesses in the AI era. While compute and models are crucial infrastructure components, data provides the true competitive moat. This sentiment is powerfully illustrated by Google’s $10 million acquisition of Spirit Airlines’ data from bankruptcy, not its physical assets, for training models. This event, alongside increasing demand from AI labs for historical enterprise datasets (e.g., Wall Street seeking hedge fund data), signals a significant shift where accumulated data, previously “sitting on a shelf,” is now considered “gold” and a key differentiator. The challenge lies in the scarcity of high-quality, real-world datasets for training agents, making existing enterprise data highly sought after for fine-tuning and developing practical applications.

Eon’s Approach to Unlocking Enterprise Data for AI

Eon has developed a new data foundation designed to map, classify, ingest, and manage an organization’s scattered data across multiple hyperscalers. The core problem Eon addresses is the inaccessibility and siloing of data within enterprises, often locked away by different business unit owners or residing in legacy systems. This makes it challenging for data teams to leverage it for AI initiatives, especially when production systems, security, and compliance are at risk.

Eon’s solution provides:

• Data Mapping and Classification: Identifying what data exists, its sensitivity, and where it resides.
• Efficient Ingestion: Easily bringing structured and unstructured data from diverse sources into a cost-effective data foundation.
• Semantic Layer: Building context on top of data, enabling intelligent querying and search.
• Access Controls and Masking: Granularly controlling access and masking sensitive information (e.g., PII, financial data) to ensure compliance and security without compromising production.
• AI Workflow Integration: Making this cleansed, contextualized data readily available for applying AI models and LLMs.

This approach transforms historical data into an accessible, valuable resource, enabling organizations to “go wild” with their data, foster a culture of “builders,” and extract previously unimaginable insights.

Emerging Security Threats: Autonomous AI Agents

A significant portion of the discussion focuses on the evolving threat landscape, where the primary concern has shifted from human threats (like ransomware) to non-human actors: autonomous AI agents. These agents can possess legitimate access and permissions within an environment, enabling rapid and potentially destructive actions, such as dropping database tables. The velocity of these incidents is “extreme,” and while detection methodologies may be similar to traditional threats, the scale and nature of the perpetrators are new.

The implications are profound for enterprise security:

• “Assume Breach” Mindset: Organizations must operate under the assumption that breaches, whether malicious or accidental, will occur.
• Non-Human Identities: The proliferation of AI agents creates “a complete set of actors inside the organization not bound by the rules of the organization,” handling sensitive data outside traditional control frameworks.
• Challenges for IT/CISOs: Leaders are caught between the board’s push to enable AI and the terrifying security implications of non-technical users deploying agents with company data without awareness of security or compliance.

This necessitates a re-evaluation of security postures, especially concerning endpoint security and the tracking of non-human identities and their chains of responsibility.

Re-imagining Data Infrastructure for the Agentic World

The rise of AI agents fundamentally alters the traditional enterprise stack, which was built for human-driven analytical questions. Old ETL pipelines, dashboards, and BI tools, while valuable, lack the unified context needed for agents that can reason over vast, disparate datasets and interact dynamically with various systems (e.g., SAP, SaaS apps, historical archives).

Key changes and requirements for new data infrastructure include:

• Unified Context: The ability to smartly collect, clean, and activate all organizational data, providing a holistic view that was previously unnecessary for single-purpose data operations.
• Scalability for AI-Generated Data: The sheer volume of data, much of it generated by agents, demands tools capable of handling immense growth and noise while extracting value.
• Cost Efficiency: Managing “tons more data” requires very cost-efficient storage and processing, especially considering token costs in LLM applications.
• Enabling Innovation: A robust data foundation empowers teams to create “wonders” by connecting previously isolated data points (e.g., combining customer preferences for pizza and burgers).
• Vendor Evolution: Even established data companies like Databricks are reinventing themselves to address the new realities of agent-generated data and the need for comprehensive data management.

The AI Era vs. The Cloud Era: A Pace on “Steroids”

Having lived through the cloud transition with CloudEndure, acquired by AWS, Ofir and Gonen draw parallels and distinctions with the current AI revolution. While cloud migration involved “thousands, tens of thousands, or hundreds of thousands of servers” and significant effort, the AI era is “on steroids” in terms of speed and disruption.

Key differences and observations:

• Velocity of Transformation: AI-driven transformations are happening “way faster” than cloud adoption.
• Loss of Control: The rapid pace leads to customers “losing control to a point where that’s becoming an inhibitor,” causing them to pause due to fears of data leaks and IP breaches.
• Ubiquitous Understanding: Unlike the cloud, which was “somewhat abstract” (“just someone else’s computer”), “everyone understands AI” due to experiences like ChatGPT. This immediate comprehension drives immense pressure from boards and CEOs to adopt AI to remain relevant.
• New Software Consumption: The AI era is fostering new models like “forward-deployed engineers” in legacy enterprises and rapid Product-Led Growth (PLG) for AI infrastructure, shrinking sales cycles.
• “Buy and Transform” Strategy: Companies are increasingly buying and transforming existing businesses into AI-driven entities to achieve higher margins and efficiency, representing a radical new way of adopting AI.

The overall sentiment is that while the journey is wild and sometimes scary, the AI revolution is only just beginning, promising a world that will ultimately be “better because of that.”

• Tags:
• Autonomous Agents
• AI Security
• Feature
• Data Infrastructure
• Ransomware
• Cloud Migration
• Data Governance
• Enterprise Data
• AI Data Strategy
• Data as Moat
• LLM Training Data
• Eon.io
• Data Classification