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

Summary

On No Priors, Eon co-founders Ofir Ehrlich and Gonen Stein describe their cloud-based data foundation as infrastructure for mapping, classifying, ingesting, protecting, and making enterprise data usable for AI. Their central thesis is that company data has become the most valuable asset in an AI world because compute and models are relatively easy to switch, while proprietary data is scarce and strategically defensible. They point to examples like Google’s purchase of Spirit Airlines’ bankruptcy data as evidence that real-world datasets now have direct training value. The conversation also focuses on security: they argue that AI agents with legitimate permissions create a new threat surface, where actions can happen faster than humans can monitor, making visibility, auditability, and governance essential. They frame Eon as a response to both the data-access problem and the emerging agentic security problem.

Notes

• Eon’s product is a cloud-based data foundation for discovering, classifying, ingesting, protecting, and querying enterprise data across multiple hyperscalers.
• Enterprise data is positioned as the primary moat in the AI era.
• Google’s reported Spirit Airlines data purchase is cited as proof that bankrupt-company datasets can have standalone strategic value.
• They say AI labs need real training data, and synthetic data often looks too synthetic to be useful.
• The guests expect more use of existing enterprise data, including sensitive data, to create real-world training sets.
• Eon’s core workflow is finding data, understanding it, classifying it, mapping context, and building a semantic layer before any AI use.
• The product is described as aggregating historical and current data, masking sensitive fields, and exposing the result to AI models.
• The company argues that data sprawl, storage cost, and token cost force teams to be selective about what they retrieve and process.
• The founders compare AI transformation to cloud migration, but say AI is moving much faster and causing operational fear.
• AI is presented as more visible than cloud, which increases executive pressure to adopt it quickly.
• They expect dashboards to proliferate because humans need visibility into chains of non-human actors.
• Nontechnical employees using tools like Lovable may unknowingly introduce security and compliance risks.
• The security model is shifting from human threats to agents that already have legitimate permissions.
• Rogue agents can create destructive incidents quickly, even without abusing credentials in an obvious way.
• The founders imply that the same detection patterns used for ransomware can be adapted to agentic threats.
• Eon is framed as a way to access scattered data across business units without breaking security, compliance, or production systems.
• They emphasize auditability as part of the control plane for AI-era data access.
• Historical and current datasets are treated as a unified source for AI applications once permissions and masking are in place.
• The guests argue that access control must be contextual, not just storage-oriented.
• They see a need for semantic understanding on top of raw data before it can be used by AI systems.
• The broader implication is that AI-era infrastructure must unify governance, retrieval, and model access rather than treat them as separate layers.
• The data stack is under pressure from both greater scale and more expensive consumption, pushing teams toward selective retrieval and efficiency.
• The guests suggest AI adoption is being driven both by upside and by fear of irrelevance.
• The pace of transformation is described as faster than the cloud era because AI is more legible to business leaders.
• The ecosystem is expected to need more governance, monitoring, and endpoint-style controls as agent use expands.
• Overall, the episode frames enterprise data management as the control point for both AI value creation and AI risk reduction.