From Raw Agentic AI to Context-Aware Enterprise Systems
Knowledge graphs AI agents for enterprises are systems where autonomous AI components operate over a structured map of business entities, relationships, and rules so they can reason over governed data access and act in line with real-world constraints, rather than generating isolated outputs from unstructured information. Agentic AI has moved from lab demos to boardroom slides, but many deployments stall because they lack this structure. Gartner predicts over 40% of agentic AI projects will be canceled, a warning that points to weak orchestration, missing domain context, and poor alignment with existing systems. Pega’s Peter van der Putten argues that “people have maybe some magical thinking that you just throw an AI model at a problem and it will sort itself out.” In practice, agents must be coordinated, supervised, and grounded in the specific business graph they operate in.
AWS Context: A Knowledge Graph Backbone for Enterprise Agentic AI
AWS Context shows how AI orchestration systems can sit on top of scattered enterprise data and turn it into a usable substrate for agents. Rather than extending data lakes and warehouses alone, AWS Context maps relationships across databases, streams, and institutional knowledge into a continuously updated knowledge graph. Mai-Lan Tomsen Bukovec describes it as a “data lake of nuance and information that AI agents swim in” so they can reason about dependencies, business rules, and domain knowledge. The service enables governed data access: teams can exclude test or sandbox datasets and set guardrails around what agents can see or act on. That matters for security and compliance, but also for accuracy, because agents can follow multi-hop chains such as vulnerabilities to systems to codebases to applications to affected users. Context becomes infrastructure, not an afterthought.
Pega: Why Orchestration Beats Raw Agentic Power
Pega’s approach highlights that enterprise agentic AI fails without deliberate orchestration layered over existing decisioning engines and customer data. The company’s Customer Engagement Studio adds a governed agentic workspace on top of its Customer Decision Hub, coordinating specialized agents for strategy, creative, data science, compliance, and performance through one conversational interface. Wells Fargo’s deployment shows what is at stake: Pega’s platform supports “six billion next best action decisions every month, across every channel, in under 250 milliseconds.” The limiting factor is not model horsepower, but the ability to feed that engine with the right content, options, and guardrails. Orchestration here means mapping agents to clear roles, embedding them in known workflows, and aligning them with measurable outcomes rather than token usage. In this view, the knowledge graph of customers, offers, and constraints is the canvas, and agents are coordinated workers on top.
Kantata’s Expertise Agent: Knowledge Graphs in Daily Service Delivery
Kantata’s Expertise Agent brings these ideas into professional services, where the connection between people, skills, projects, and margins defines success. The agent sits on a services-native knowledge graph that joins data from projects, resources, systems, documents, communications, and meetings, then maps how skills, delivery patterns, and outcomes relate. Kantata reports that 87% of professional services organizations plan to use AI agents as part of their workforce, while 89% of leaders say future revenue growth will depend more on scaling AI than scaling headcount. Expertise Agent aims to close the gap between insight and execution: it interprets complex questions, orchestrates actions across PSA workflows and external tools, and can create other agents when needed. That enables resource optimization, early risk detection, and preservation of institutional knowledge during handoffs, moving beyond dashboards to agents that act with operational context.
Why Knowledge Graphs and Governance Decide Enterprise AI’s Future
Across AWS, Pega, and Kantata, a pattern is clear: enterprise agentic AI succeeds when knowledge graphs and orchestration come first. Knowledge graphs give agents a shared model of business context: who and what exists, how they relate, and which constraints apply. Orchestration then coordinates multiple specialized agents across that model, aligning them with outcomes and compliance requirements. This combination turns raw models into AI orchestration systems that can handle real workloads such as customer engagement, professional services delivery, and cross-system operations. It also supports governed data access, letting teams decide which sources and relationships agents can inspect or act on. Vendors that treat knowledge graphs as core infrastructure, not optional metadata, are better positioned to build agents that understand work, capacity, risk, and profitability, instead of generic automation that cannot explain or justify its actions.
