Mphasis is urging enterprise technology leaders to stop treating their legacy core systems as “untouchable” as they rush to adopt agentic AI. Instead, the company argues, AI needs to be built into the architecture itself, with an explicit semantic layer that explains how the business really works. Without that, powerful agents will simply automate and amplify the flaws, inconsistencies, and blind spots already buried in core systems.
At the centre of this push is Mphasis’ NeoIP platform and its Ontosphere layer, described as an ontology-driven enterprise knowledge graph. Rather than being just another data catalogue, Ontosphere attempts to capture business rules, code paths, operational workflows, documentation, and incident history into a single structured model. For CTOs, the pitch is clear: if you want agentic AI to safely orchestrate systems end-to-end, it first needs a trusted “map” of what your concepts mean, how they relate, and where logic actually lives.
Ontology as the Control Layer for Agentic AI
Research firm HFS backs this view, identifying ontologies and knowledge graphs as a critical control layer for next-generation AI. Agentic systems can already trigger workflows, call APIs, and take decisions at scale. The risk, HFS cautions, is that when they are pointed at legacy estates with fragmented rules and undocumented exceptions, they can propagate incorrect intent or recreate outdated policy in new digital channels.
Ontosphere is Mphasis’ answer to that problem. By continuously extracting and reconciling intelligence from code, tickets, documents, and processes, it creates a persistent semantic layer that agents can query and rely on. Instead of learning the enterprise “from scratch” on raw logs, AI systems are grounded in a graph that encodes what a customer, policy, claim, transaction, or risk actually means in that specific organisation.
From Incremental Modernisation to Architectural Rethink
Mphasis notes that, across financial services, insurance, and the public sector, most modernisation has been incremental: cloud migrations, new digital front-ends, API wrappers—while the core business logic remains largely unchanged. That pattern was sustainable when AI lived at the edge, powering chatbots or analytics dashboards. With agentic AI now being asked to ingest applications, automate decisions, and orchestrate entire journeys, the quality of the underlying logic becomes mission‑critical.
Nitin Rakesh, CEO and MD of Mphasis, frames this as an architectural turning point. AI, he argues, cannot be treated as a thin layer on top of fragmented systems; intelligence must become part of the architecture itself. Without that shift, each new agent adds complexity rather than clarity.
Early Results: Incident Prediction and Faster Recovery
To make the case concrete, Mphasis cites results from a global insurance client using its ontology and knowledge graph framework in IT operations and observability. By aligning events, services, and dependencies in Ontosphere, the insurer reportedly achieved around 67% accuracy in predicting major incidents, gained three to five hours of early warning, and cut mean time to detect, acknowledge, and resolve incidents by about 50%. For AI leaders, this is a template: when the system understands how components relate, proactive operations become possible.
A “System of Truth” for Enterprise Intelligence
HFS executive research leader David Cushman argues that enterprises don’t lack intelligence—they lack a way to consolidate it. Critical know‑how is scattered across source code, SOPs, emails, and legacy tools. Ontologies and knowledge graphs, he says, provide a “system of truth” for that intelligence, making it reusable for AI without repeating legacy mistakes.
As agentic AI matures, both Mphasis and HFS expect architectural discipline to matter as much as model choice. The organisations that succeed will be those able to codify their own institutional meaning and then let AI operate within that well‑governed semantic boundary, rather than unleashing powerful agents on opaque, aging cores.
