2025 was supposed to be the year of the AI agent. And such agents, which can handle everything from basic research and coding tasks, are indeed popping up more and more.
But why merely replicate human workflows when AI can reinvent them entirely? That’s part of the philosophy driving illumex CEO Inna Tokarev Sela’s vision for enterprise AI—one where agents don’t just follow existing processes but recombine experiences across environments. This vision starts with creating what Sela calls an “organizational single source of truth” that embeds business logic, data structures, and even user interaction patterns into a knowledge graph with semantic embeddings. She describes this agent-driven approach as “almost like quantum reality” where systems can test multiple possibilities simultaneously instead of following linear processes. Rather than just automating existing workflows, Sela believes we’re rapidly approaching “an application-free future” where AI agents fundamentally transform how work happens. She sees this technological shift as being “as significant as the Internet” with the key difference being that business leaders, not just technologists, are driving adoption.
But to realize such an ambition won’t happen if you suspect your AI might not be always telling the truth. Or if it can’t understand your business and what you are trying to achieve in the first place. It’s fitting that the company’s tagline on its website is this: “Get your data speaking the way your employees do.”
Multiple versions of truth are unacceptable
In order to get there, organizations need to develop trustworthy AI. “You cannot allow multiple versions of truth, because then you have zero trust in results when users start asking questions,” Sela said.
Inna Tokarev Sela
Inna Tokarev Sela
By now, it is easy to see the potential of AI. But also its pitfalls. What makes this current wave of attention surrounding AI different from past IT trends, Sela believes, is that “the pull for this technology is coming from the business side.” Unlike previous innovations that IT departments had to advocate for, “every executive, chief digital officer, and CEO has a bunch of ideas about what they can do with this technology. I think this wave is as significant as the internet in the sense that business people are inventing the ultimate use cases.”
In an interview at NVIDIA’s GTC conference, where illumex touted a partnership with the GPU giant, Sela explained how its technology helps R&D-heavy organizations like pharmaceutical manufacturer Teva overcome genAI’s hallucination problem. “What we’ve done is create not only industry context, which we have, but also automatically create organizational context,” she said. By embedding business logic, data structures, and even user interaction patterns into a knowledge graph with semantic embeddings, illumex creates a foundation for trustworthy agents that can operate beyond simple task automation. For R&D professionals and enterprise organizations more generally, this means AI that understands the specific context of their work rather than generating responses based on generic training data.
To power this capability, illumex has integrated key NVIDIA NeMo components including Curator for ontology metadata, Retriever for capturing user-specific patterns, and LLM NIM for context extraction and reasoning automation.
To bridge that gap between generic AI and highly specialized enterprise workflows, illumex works from the ground up to construct a single source of truth within each organization. The key, she explained, lies in embedding business logic, data structures, and user interactions into a knowledge graph enriched by semantic embeddings. “We connect to systems, pull out usage logs, usage traces, and embed this knowledge of what users usually interact with,” said Sela.
AI that can show its work
Sela’s approach to enterprise AI rests on three foundations: AI-ready data with quality and governance, deep contextual understanding through semantic models, and perhaps most importantly, “empathy toward business users.” As she puts it, “There’s not going to be any adoption if users don’t understand how the answer was calculated.”
This philosophy delivered tangible results in a recent case study where illumex helped a major retailer migrate its ERP system to cloud-based S/4 HANA. With over 115,000 tables and 1.5 million data columns at stake, the platform “transformed the otherwise two-year manual data review into a single week of automated workflows,” while translating technical data into natural language that business leaders could readily understand.
These capabilities address significant market concerns. According to illumex’s research, 35% of organizations cite AI mistakes with real-world consequences as their top adoption barrier. This explains why less than 40% of today’s workforce uses AI tools, despite 78% of organizations planning to increase AI spending in 2025, according to Deloitte.
To broaden adoption, illumex recently integrated its Omni platform into Microsoft Teams, offering an “intelligent conversational interface” for accessing enterprise data using plain English. The company promises responses that are “deterministic, hallucination-free, fully explainable, and traceable.”