Mordor Intelligence estimated the size of the global 2025 agentic orchestration and memory systems market at $6.27 billion, predicting that market would reach $28.45 billion in 2030.

That’s a considerable market, especially considering that most software companies only began releasing agentic orchestration in 2025. Adobe, for example, released Adobe Experience Platform Agent Orchestrator for Businesses, which targeted customer experiences and marketing workflows, in March 2025.

How can such a market start large to begin with and grow so fast? Because agent orchestration as incumbent software vendors define it is just workflow optimization redefined so that agents are the main focus. Enterprise software vendors have spent decades studying and automating the workflows of their customers. They’re able to build on that knowledge simply by shifting to an agent-oriented development paradigm.

The problem with this mainstream approach to agentic orchestration is that managing agents safely and effectively requires a very broad knowledge foundation that most enterprises don’t have. Agents just don’t have the context they need to make wide-ranging decisions.

Enterprises have to work at the data layer, with a data-centric, knowledge-rich approach to fix that problem. This diagram, based on Semantic Arts’ observations about application centricity versus AI ready data centricity, underscores the nature of the Before and After states enterprises at the heart of AI-ready enterprise transformation. Process optimization in the After state is part and parcel of data-layer transformation.

In order to focus on the data layer and integrate and interoperate with agents at that layer where the agents can operate effectively, they’ll have to de-invest in the application layer that’s just getting in the way of their AI-ready transformation efforts.

Essentially, enterprises have a self-destructive software buying habit they’ve developed over decades. They’ll have to break that habit and spend much less on what leading software vendors are selling them to transform their process optimization methods focused on the data layer. The savings from application layer software cutbacks can go directly to fund the creation of AI-ready data, data that’s enriched with structured contextualizing knowledge.

This post will explore the myths that surround this counterproductive approach to agent orchestration, and what the AI-ready, data-centric alternative is.

Myth #1: Leading software vendors use best practices and care about you.

The myth perpetuated every day is that leading software vendors have your best interests at heart. The reality is that they don’t.

Leading software vendors can’t help themselves. They urgently want to protect the installed base of underutilized, application layer code that they’re selling. They even want to add to that installed base, even in the AI era.

The legacy application layer these vendors favor is where all the code sprawl, duplication and data siloing originates. To work with agents effectively, enterprises should focus squarely on the data layer and the semantic metadata. That metadata must include logically connected taxonomy and ontology models.

The future of AI-ready data is in logically self-connecting knowledge graphs as your data-centric foundation.

Most software vendors keep you at the application layer where the bad habits are.

Most software vendors are like consumer packaged goods companies that dominate the center aisles of grocery stores. They’re not interested in selling nutritious items, but want you to buy processed foods instead.

Similarly, most software vendors want to keep you buying application-centric software. The more you do that, the more you’ll just perpetuate your legacy habit, buying a bunch of code that proliferates siloed data repositories. That duplicative code just gets in the way of ultimate AI effectiveness and efficiency, which is in semantic metadata at the data layer, less code sprawl and data siloing.

Myth #2: The AI we have is the AI we need.

I’ve written before about monolithic AI and the tendency for most data science efforts to focus merely on statistical machine learning, rather than leveraging a combination of neural net and symbolic AI. Scientists call such a blend NeSy, short for “neurosymbolic AI.”

But in addition to NeSy approaches such as GraphRAG that allow LLMs to query and retrieve facts from databases, some researchers are also exploring blends of human + algorithm learning they say are more effective than algorithmic learning alone.
In “Effective Generative AI: The Human-Algorithm Centaur,” Soroush Saghafian and Levon Idansay that AI works best not by replacing humans, but by combining human intuition with machine precision — a blend of human and machine learning they refer to as “symbiotic learning” and call the “centaur” model. Here’s how the centaur model works:

_{From Soroush Saghafian and Levon Idan, “Effective Generative AI: The Human-Algorithm Centaur,” Harvard Data Science Review, no. Special Issue 5 (2024), https://doi.org/10.1162/99608f92.19d78478.}

Myth #3: Packaged agent orchestration is something new, different and essential.

I looked at Gartner’s short list of agentic orchestration vendors. Despite Gartner’s recommendations for outside-in integration and knowledge graphs, the list is a mix of two groups: vendors who have simply repackaged old workflow optimization tools and monolithic AI companies who extend their agent model to become a multi-agent system (MAS) with their own proprietary management or “orchestration” layer.

Gartner recommends reverting to older workflow optimization or Robotic Process Automation (RPA) when agents are overkill. RPA vendors like UIPath, now in agentic orchestration, continue to cover core workflow optimization. That’s despite the fact that most companies who’ve tried RPA have had an unpleasant experience trying to use it. And it’s a tactical solution that doesn’t solve the problem of how to create an AI-ready data environment.

Dwelling more at the application layer to fix workflow isn’t the answer. To be AI-ready, enterprises need to create an AI-ready data layer with tiers of contextualized knowledge that maps to instance data.

But I still don’t believe their orchestration abilities extend to all the layers of contextualization of standard knowledge graph/GraphRAG providers.

Myth #4: Companies can keep their old architectures in the AI era.

The old habit 80+ percent of large enterprises seem unable to break is to spend most of their IT budget on application software year after year based on an outdated, application-centric architecture. Granted, that software tends to be familiar, and it’s a straightforward process for IT to implement and train the workforce once subscribed to software as a service.

But companies are kicking the can down the road by paying for packaged AI-capabilities that come with the suites they subscribe to.

A better approach: Get into the habit of talking to and working with your graph.

As I’ve pointed out before, enterprise IT suffers from various kinds of complexity and duplication, including code bloat and data siloing. The proliferation of redundant applications and fragmented data repositories is a primary problem that needs to be resolved by breaking old habits. Software buying habits in place since the 1980s are responsible for adding to the complexity and duplication.

This software wasteland—credit to Dave McComb for his 2018 book on the topic—reflects an application-centric mindset that prioritizes siloed systems over shared data. Instead of adopting a decentralized, web-like architecture, most enterprises because of inertia perpetuate and even add to application-centric technical debt by subscribing to more services.

GraphRAG and a semantic backbone harmonizing these disparate sources into a unified knowledge graph, shifting the focus from maintaining bloated software to managing relationship-rich, interoperable data–the data-centric approach McComb describes in his book.

Ultimately, to escape the software wasteland and become AI-ready, enterprises must overcome decades of inertia and fundamentally shift their focus. The solution lies in abandoning the application-centric model that perpetuates technical debt, duplication, and siloed data.

By adopting a data-centric architecture anchored by a unified knowledge graph and a semantic backbone, companies can resolve complexity, manage relationship-rich, interoperable data, and build the AI capabilities they really want and need, rather than just buying more packaged goods that don’t help much.

For more information:

Adobe. “Adobe Launches Adobe Experience Platform Agent Orchestrator for Businesses to Activate AI Agents in Customer Experiences and Marketing Workflows.” News release, March 18, 2025. https://news.adobe.com/news/2025/03/adobe-launches-adobe-experience-platform-agent-orchestrator-for-businesses.

McComb, Dave. Software Wasteland: Why the Cost of IT Keeps Going Up and What to Do About It. Basking Ridge, NJ: Technics Publications, 2018. https://technicspub.com/software_wasteland/.

Mordor Intelligence. Agentic AI Orchestration and Memory Systems Market Size, Share & 2030 Growth Trends Report. July 2025. https://www.mordorintelligence.com/industry-reports/agentic-artificial-intelligence-orchestration-and-memory-systems-market.

Morrison, Alan. “Aaron Philipp: How Graphs and Semantics Boost Cybersecurity Potential.” The GraphRAG Curator, December 1, 2025. https://graphrag.info/2025/12/01/aaron-philipp-how-graphs-and-semantics-boost-cybersecurity-potential/.

Morrison, Alan. “Contextual GraphRAG and Its Evolution.” The GraphRAG Curator, February 11, 2026. https://graphrag.info/2026/02/11/contextual-graphrag-and-its-evolution/.

Morrison, Alan. “More on AI and Semantic Layer Contrarians versus Conformists.” The GraphRAG Curator, November 19, 2025.https://graphrag.info/2025/11/19/more-on-ai-and-semantic-layer-contrarians-versus-conformists/.

Saghafian, S., and L. Idan. “Effective Generative AI: The Human-Algorithm Centaur.” Harvard Data Science Review, no. Special Issue 5 (2024).https://doi.org/10.1162/99608f92.19d78478.