The Pitch Says Agent. What Does the Tech Actually Prove?

Walk into any enterprise software room right now, and you’ll hear the same buzzword echoed from every corner: Agentic.

Whether it is CRM, expense tracking, or enterprise content management, it seems every legacy system has suddenly become an "AI agent" overnight. But as our CEO, Eric Barroca, highlights in his recent deep-dive, we are currently living through a major case of "agent-washing" - where marketing claims are moving much faster than actual product architecture.

Enterprise content is arguably the single largest opportunity for AI today. For thirty years, content management platforms promised to truly organize, classify, and make business data "findable," only to pass the actual burden of understanding that data back to humans.

Now, we finally have the technology to change that. But instead of rewriting the rules, many software vendors are simply slapping an "agent" label on old tech.

So, how do you separate the marketing spectacle from substance?

The ultimate test: read the API, not the slides

An actual AI agent doesn't just respond to a single prompt or follow a rigid, pre-programmed human workflow graph. An agent reasons. It takes a goal, decides its next move, selects the right tool, reviews the result, and loops until the job is done.

If you want to know if a platform is genuinely agentic, you shouldn't look at their marketing deck.

You have to look at their code interface (the API). A true agentic platform leaves a clear trail of its reasoning loop. If it doesn't, it’s not an agent; it’s just a glorified chatbot or a legacy workflow masquerading as one.

When evaluating your tech stack, ask four fundamental questions:

1. Does anything actually reason? Is there a dynamic "loop" where the AI thinks through intermediate steps, or is it just answering a one-off question?
2. Is the system reading the content or just indexing text? True agents need to understand a document as a document - interpreting things like layout, signatures, clauses, and tables - not just flat strings of text.
3. Is the data structure fixed or alive? Are you still forcing AI to fill out a template drawn up by a human decades ago, or is the AI organically structuring the information based on the content itself?
4. Is search just a tool, or is it the entire product? Simply slapping "vector search" (RAG) onto an existing document store doesn't make it an agent runtime.

Redefining the core tech: Primitives, Repositories, and Process Engines

To understand why simply adding an "AI layer" to old software fails, we have to look at how software is built. Real innovation requires changing the foundational architecture. In enterprise content, this comes down to three core concepts:

• Primitives: In technology, a primitive is the lowest-level, fundamental building block from which a system is constructed. For decades, the primitives of content management were files, folders, and static checkboxes. True AI requires a shift to a new primitive: a dynamic, "document-native context layer" that an AI can navigate natively.
• The Repository: A repository is the centralized digital vault where documents, data, and access controls live. Legacy repositories simply hold a file and extract text for human keyword searching. An AI-native repository preserves relationships, tables, and visual layouts so an AI agent can read and interpret a contract just like a human expert would.
• The Process Engine: A process engine is the "traffic cop" of a software platform - the backend orchestrator that routes tasks, checks business rules, and executes workflows. Traditional engines rely on a rigid graph drawn in advance by a human (e.g., Step A must always lead to Step B). An agentic process engine acts as a fluid contract where AI agents, tools, and humans can collaborate dynamically, allowing the AI to determine its own next steps while strictly enforcing legal and operational guardrails.

AI truly changes the primitives

At Vertesia, we believe AI isn't just an extra feature layer to bolt onto a legacy system. It completely shifts these fundamental building blocks.

Adding an AI label to an old repository or a rigid workflow engine is like putting a fresh coat of paint on a crumbling foundation. To truly unlock the promise of AI for enterprise content, we have to rebuild the repository and the process engine from first principles so they are built natively for machines that can reason.

The tech industry has finally been handed the machine it always wanted. Let’s not waste the moment selling paint.

Want to dive into the technical details? Read Eric's breakdown of major legacy platforms on Substack: Enterprise Content Finally Has the Machine It Needed. Don't Wash It Away.