Enterprise AI adoption faces a fundamental constraint: the data that powers intelligent systems remains locked behind legacy applications, proprietary software, and interfaces that were never designed for modern integration. While traditional tools have connected enterprise systems for years, they fall short of delivering the complete information and business context that AI agents require to function effectively.

Ran Cherniavsky, CEO and Co-Founder of Unfold, has built his company around a different premise. Rather than replacing existing infrastructure, Unfold works through the interfaces that businesses already operate—extracting data and mapping the underlying logic without requiring system overhauls or disruptive migrations. In conversation, Cherniavsky outlines why data accessibility has become the defining challenge for enterprise AI, how his platform unlocks information from otherwise sealed systems, and what becomes possible when AI agents gain visibility into an organization's complete operational picture.

The enterprise data problem

Integration technology was designed to connect systems to one another, not to serve artificial intelligence. Traditional approaches link system A to system B by exposing whatever fields vendors chose to make available, one at a time. AI, by contrast, demands access to everything a human operator can see within a system, along with an understanding of what that information means. This must happen quickly, without launching a separate project for each data field. The problem integration was built to solve differs fundamentally from what AI requires.

The consequences of leaving data inaccessible extend across multiple dimensions. Organizations that cannot deploy AI on their core datasets will fall behind competitors who can, and this gap widens rapidly. Before that competitive disadvantage materializes, companies incur substantial costs: they hire specialized consultants year after year to manually map system fields, analysts generate capped reports and stitch them together in spreadsheets, and AI pilots stall because models only see the narrow slice that an API exposes, producing incorrect results. In essence, companies rent access to their own data through someone else's interface.

An invisible risk compounds these visible costs. When nobody can read what resides inside a system, nobody can confidently assert that it remains secure. Organizations are left hoping their data stays protected.

Among security concerns, IT modernization initiatives, and AI adoption pressures, inaccessible enterprise data creates its most urgent problem in the AI adoption category—and the gap is substantial. Board-level pressure focuses on AI deployment, and locked data imposes immediate costs there. A dealership owner cannot ask an AI assistant how much warranty money is owed to them because that information sits trapped behind reports. The question itself is straightforward to phrase in plain English; the barrier is data accessibility, not language.

Security and modernization face the same underlying issue, though it manifests differently. Security becomes a symptom: organizations cannot secure what they cannot see. Modernization represents a lengthy workaround. According to Gartner, 45% of core systems still operate on mainframes or on-premises client/server architecture, and 34% of organizations now choose to leave core applications unchanged because the risk of modernizing them is too high. AI cannot wait for that transformation to complete. Solving data access and context for AI simultaneously addresses significant portions of both the security and modernization challenges.

Unfold's technology and differentiation

Unfold accesses systems lacking APIs, exports, or open interfaces by mimicking how employees actually work. Staff members do not interact with systems exclusively through APIs; they navigate user interfaces, generate reports, export files, and use multiple access methods. Unfold operates through every interface an application exposes, whether or not an API exists.

The company's team conducted years of deep software research to integrate with systems regardless of what they made publicly available. This work historically required months of manual effort per system. Using AI, Unfold now becomes a subject-matter expert on any system within days, identifies where every data element lives and how it flows, then constructs a deterministic read-only collector that extracts it through the system's existing permission structure.

This approach functions on mainframes like the AS/400, on-premises ERPs, and modern SaaS platforms that expose only partial data. If a system has a user interface or command line, Unfold can work with it. The onboarding process takes up to 14 days for any system.

Raw data divorced from meaning becomes noise. In Oracle's ERP JD Edwards, database tables contain numbers rather than descriptive names. Status fields hold codes whose original definitions nobody recalls. Two different systems refer to the same customer using different identifiers. Unfold learns each system the way a trained employee would: understanding what each object represents, how objects connect to one another, which business rules and exceptions apply, and how people actually use the system in practice. Because Unfold observes both the screen display and the underlying data, it can connect what users see in the interface to where that information resides in the database.

This knowledge traditionally lived with subject-matter experts, implementation partners, and vendors—much of it departing when those people left. Unfold reconstructs it as a comprehensive catalog documenting every data object, its business meaning, and its complete journey through the system. AI receives both the data and its context simultaneously, eliminating guesswork.

Once each system is understood, Unfold connects them together. The same customer appears as KUNNR in SAP and as an Account in Salesforce. The same location carries different IDs across each ERP. Unfold maps all of these into a single canonical enterprise schema, ensuring data from every system arrives normalized and consistent.

Governance is embedded from the beginning. Data remains with the customer in their own source systems. Lineage accompanies it, since Unfold tracks where every field originated and how it was transformed. The collectors operate deterministically: the same query returns identical data every time. This consistency is essential before allowing an agent to act on the information.

Unfold complements rather than replaces existing platforms like Splunk, Snowflake, and Databricks. These platforms serve as destinations where data gets searched, analyzed, and activated by AI—they are only as effective as the information that reaches them. Currently, that means whatever a source system's API or export permits. Unfold can alternatively work with data where it lives, in the source systems themselves, or deliver it directly to agents through connectors to Claude, Gemini, and other platforms.

Unfold operates upstream in the data pipeline. It extracts data and logic from systems that never fed traditional data platforms, preserving business meaning, and delivers it wherever customers need it: their data lake, their warehouse, or their agents directly. At a global retailer, point-of-sale data from an AS/400 mainframe now flows into a Gemini Enterprise fraud-detection agent. At a manufacturer, plant control system data feeds into a CMDB.

Traditional integration begins with a catalog of prebuilt connectors and moves whatever each vendor exposes, one project at a time. Unfold starts from the system itself: every interface, the complete data model, and the business context, accomplished in days rather than months.

Customer applications and business impact

The most compelling use cases follow a consistent pattern: someone observes data displayed on a screen but cannot move it into the workflow they need. Real examples illustrate the impact:

  • A global retailer requires point-of-sale data from AS/400 systems across more than ten countries to feed its fraud detection system, which now runs on Gemini Enterprise. A regulator in one country demanded the company halt sales until this control was implemented.
  • A top-10 US insurance company employs contractors with specialized expertise to manually map claims-system fields to the database. A separate engineering team then builds the ETL pipeline. The entire process takes weeks. Unfold accomplishes it automatically in days.
  • A multi-store auto dealership group had its parts, service, and warranty data locked behind reports in its dealership management system.

Now the owner asks, in plain English, "How old is the warranty money owed to us?"

Customers measure value through three dimensions: time (days instead of months), cost (elimination of consultants, professional services, and engineering resources), and new outcomes that were previously impossible, such as answers spanning multiple systems or business continuity that would have required a halt to operations.

Unfold enables enterprises to modernize legacy infrastructure without disrupting critical operations or replacing existing systems. It accomplishes this by not touching the systems at all. Unfold requires no code changes or migration; it works with systems as they exist, through interfaces already in place, and operates in read-only mode by design.

This non-invasive approach matters because modernizing core systems carries genuine risk, and organizations understand this clearly. Gartner research shows that 34% of organizations now leave core applications unchanged for exactly this reason, up from 16% in 2023, and that 40% of the modernizations required for autonomous business will be delayed by 2030.

Rather than waiting years for a system replacement, organizations can extract data and business logic today and deploy AI on top of it immediately. When modernization eventually occurs, they already possess a complete map of what the legacy system contains and what it means—a substantial portion of the migration effort.

When AI agents gain access to and comprehension of data from previously closed enterprise systems, AI adoption accelerates dramatically. Today, enterprises work for AI: teams invest months on integrations simply to supply data to models. Once every system is open and understood, this dynamic reverses. Organizations work with AI, and AI handles integrations and executes workflows independently.

Agents stop operating on incomplete information and start addressing actual business problems. Most agents today reach email, documents, and a handful of SaaS APIs. The business core—orders, claims, inventory, warranty, and plant operations—resides elsewhere. Opening that access transforms possibilities: a dealership owner poses a single question spanning parts, service, and warranty; a fraud model sees point-of-sale data from every country; a finance team retrieves on-time-in-full metrics across two ERPs that use different site identifiers.

The impact compounds further. Agents follow processes end to end, like order-to-cash, across every system involved, rather than stopping at one system's boundary. At that point, AI transitions from answering questions to operating the business.

The future of enterprise AI

Data accessibility has already become one of the biggest barriers to enterprise AI adoption. Gartner identifies data access and control as the largest compounding challenge for agentic AI through 2030. Models continue to improve in capability, but intelligence alone will not drive enterprise transformation. Models require access to an organization's proprietary data and understanding of how each system is used.

Organizations can overcome this barrier by beginning with business outcomes rather than data projects. They should identify one AI initiative blocked by a system from which data cannot be extracted, solve that problem, then expand system by system on the same layer. Treating data access as a series of one-off integrations should end, and organizations should not wait for modernization to complete. Additionally, the knowledge embedded in systems should be treated as an asset. The people who understood those systems are retiring or have departed. Capturing that context now is part of becoming AI-ready.

The enterprise integration market will likely follow a historical pattern over the next three to five years. First, something is invented. Then a commercial application emerges. Only years later does it fundamentally change how organizations operate. The first modern automobile appeared in 1886, but decades passed before it transformed society. That intervening time went into building supporting infrastructure: roads, gas stations, bridges.

AI occupies that same position today. Probably the most powerful engine ever built exists, and models keep improving. Yet transformation inside enterprises has not occurred, and the reason is not insufficient intelligence. The roads do not exist.

Integration represents a major component of those roads, though it was constructed for different traffic patterns. It connected systems to systems using connectors and pipes, moving one field at a time. Over the next three to five years, the traffic changes. AI becomes the entity needing the data, and it needs all of it, along with its meaning.

The market will likely split. Moving data becomes a commodity, like a basic road. Value shifts to the layer that understands the enterprise: complete access to every system, the business context behind it, and one schema across all of them.

Gartner already treats AI context platforms as a distinct market, projected to grow from $28B in 2026 to $78B by 2030. The research also notes that these platforms face limitations in how much they can actually extract from systems of record. That gap is exactly where Unfold operates.

The difference this time is that organizations do not have to wait 40 years. With AI, enterprises can build the roads much faster than they built them for automobiles, but only if their data can actually reach the engine. Unfold's goal is to lead this category and become the knowledge foundation for enterprise AI, so that every AI use case in the enterprise runs on Unfold, from the mainframe to SaaS.

The founder's vision

Cherniavsky's team spent years conducting deep software research, extracting data from systems never designed to release it. When AI arrived, suddenly every company faced the same problem his team had been solving manually, at a scale no organization could staff with people. Twenty years ago, data lived on a company's own servers. Today it sits behind vendors, interfaces, and knowledge that is retiring.

Unfold was built so AI can understand any enterprise system the way a long-tenured employee does. The long-term vision is straightforward: every AI use case in the enterprise runs on Unfold, from the mainframe to SaaS.

Source: The Next Web