Executive Summary
The artificial intelligence landscape is undergoing a fundamental shift. At NeurIPS 2024, Ilya Sutskever made a profound observation that resonates deeply with the future of AI: “Pre-training as we know it will end.” This isn’t just another AI prediction – we’ve reached “peak data,” with only “one internet” worth of training material available. This limitation is catalyzing a crucial evolution: the rise of reasoning AI agents that emphasize structured knowledge over raw data volume.
This transformation is particularly significant for the service economy. While traditional software has focused on enhancing productivity, AI agents are poised to take over service delivery itself, expanding the total addressable market from mere productivity enhancement to comprehensive service execution – a shift that opens up opportunities worth over $10 trillion.
To further illustrate the vast potential of this transformation, it helps to consider the emerging ‘application category layer’ for enterprises. Recent analysis highlights distinct areas where practical value from AI agents is being realized. These include categories like AI Assistants, directly mirroring the vision of agents like TrenIA (introduced below) transforming customer interactions, and Content and Media Generation, where agents like Agent WordLift are already demonstrating impact.
Through advances in knowledge graphs and GraphRAG technology, organizations can navigate this transition by focusing on their “ontological core” – the fundamental concepts and relationships that define their domain. This approach has shown a 30% improvement in reasoning capabilities compared to traditional approaches, as demonstrated through real-world applications in different application categories (here is a deep dive on the application category layer for enterprises).
The End of the Pre-training Era
With just “one internet” worth of training material available, the future of AI lies not in accumulating more data, but in better understanding and structuring the knowledge we already have. This stark reality signals a fundamental shift in how AI must evolve, pushing the industry toward systems that can reason, plan, and execute with existing knowledge rather than waiting for more training data.
This limitation isn’t just a constraint – it’s an opportunity to fundamentally rethink how we approach artificial intelligence. The focus must shift from raw data accumulation to knowledge structuring and reasoning capabilities. This evolution marks the beginning of a new era in AI, where the quality of knowledge organization and the sophistication of reasoning mechanisms become the primary drivers of advancement.
The Power of the Ontological Core
As Tony Seale emphasized during his keynote at Connected Data 2024, organizations need to “find their ontological core” – the fundamental concepts and relationships that define their domain. This isn’t just about organizing data; it’s about creating a resilient foundation for AI systems.
With the era of unlimited training data coming to an end, organizations must focus on what Sutskever calls the “fossil fuel of AI” – their existing data and knowledge structures. At WordLift, we’ve discovered that the key to building effective AI agents lies in this ontological core – the structured representation of an organization’s domain knowledge, relationships, and business logic.
Our SEO agent demonstrates this approach in action. By leveraging semantic technologies and knowledge graphs, we’ve created a system that doesn’t just process content – it understands it within the broader context of SEO strategy and business objectives. This has enabled teams worldwide to scale their content marketing efforts while maintaining strategic control and oversight.
TrenIA: Exploring the Future of AI Agents in Transportation
While developing Agent WordLift for SEO and content optimization, we began exploring how our agentic technologies could extend beyond digital visibility to transform customer relationships. This led to an experimental project in WordLift’s lab: TrenIA, a conceptual AI agent for Trenitalia that demonstrates how ontological cores and knowledge graphs could revolutionize service delivery for Italy’s primary train operator.
The concept emerged from a simple question: How could we apply our expertise in knowledge graphs and AI agents to enhance real-world customer experiences? By integrating WordLift’s Knowledge Graph architecture with viaggiatreno APIs, we envisioned how an AI agent could transform static train schedules into dynamic, user-focused experiences.
This exploration wasn’t conducted in a vacuum. Our sister company RedLink‘s previous work with Deutsche Bahn on “Reisebuddy” provided valuable insights into how AI can augment transportation services. Their experience, started 8 years ago (!!!), reinforced a crucial lesson: AI in customer service must be human-led, with the ontological core centered on passengers, their transportation needs, and the infrastructure that serves them.
In designing TrenIA’s framework, we identified the following components of its ontological core:
- Passengers and their journey patterns
- Transportation infrastructure (trains, stations, buses)
- Service networks and partnerships
- Real-time operational data
- Historical performance metrics
- Regulamentary frameworks
This structured knowledge foundation would enable capabilities such as:
- Real-time updates on delayed Frecciarossa and IC trains
- Dynamic route alternatives during disruptions
- Context-aware assistance based on passenger history
- Automated handling of routine queries
- Proactive travel support
From Experiment to Innovation
While TrenIA remains a laboratory experiment, it represents something far more significant: a practical exploration of how AI agents could transform service delivery. This experimental work has validated my belief that the same technologies we use to enhance digital visibility can be adapted to revolutionize customer relationships across various industries.
The insights gained from this conceptual exercise align with the broader trend of AI agents “eating” the service economy. As we’ve seen with Agent WordLift in the SEO domain, AI agents are moving beyond simple automation to become intelligent collaborators in service delivery. The TrenIA experiment demonstrates how this transformation could extend into traditional service industries, bridging the gap between complex infrastructures and user needs.
Why This Matters: The Service Economy Revolution
While traditional software has focused on enhancing productivity, AI agents are poised to take over the work itself. This isn’t just an incremental change – it’s expanding the total addressable market (TAM) for software from productivity enhancement to the actual delivery of services.
At WordLift, we’ve seen this evolution firsthand through our AI SEO agent, launched in April 2024. Our experience shows that success doesn’t just come from more data – it comes from building systems that can reason about and act upon structured knowledge. This AI-powered SEO tool automates tasks, generates high-quality content, and provides data-driven insights, demonstrably saving content and SEO managers 40-60% of their time. Furthermore, it empowers multilingual content creation and facilitates interactive customer engagement. The agent’s rapidly growing adoption (as shown by our cumulative consumption metrics below) demonstrates that organizations are ready for this transition when given the right tools and ontological foundation.
The enterprise market validates this transformation:
- Foundation model spending is projected to reach $6.5B in 2024, up from $2.3B in 2023
- Application layer investments are increasing to $4.6B from $0.6B
- AI infrastructure spending is growing to $2.7B from $1.2B
This dramatic increase reflects three fundamental shifts:
- From Tools to Agents: Organizations are moving beyond AI automation toward systems that can autonomously perform complex tasks
- From Data to Knowledge: Success depends not on raw data volume but on well-structured, semantically-rich knowledge bases
- From Silos to Services: AI agents are breaking down traditional software categories, creating new service-oriented business models
As Anthony Alcaraz from AWS put it at Connected Data 2024, we’re seeing “RPA on steroids” – but this undersells the transformation. These aren’t just automated workflows; they’re intelligent agents capable of understanding context, making decisions, and adapting to new situations.
Building Production-Ready Agentic Systems
At Connected Data 2024, Anthony Alcaraz presented a compelling framework for implementing AI agents that revolutionizes how we approach system architecture. His framework demonstrates how graph-based systems enable AI capabilities across three critical dimensions:
Strategic Intelligence:
- Enabling high-level decision making and planning through ontological understanding
- Leveraging knowledge graphs for strategic insights
- Connecting business objectives to execution capabilities
Tactical Execution:
- Converting strategic insights into actionable workflows
- Orchestrating real-time responses to changing conditions
- Managing resource allocation and task prioritization
Interface Intelligence:
- Providing seamless interaction between users, data, and AI systems
- Adapting communication based on user context
- Maintaining consistency across interaction channels
This three-layered approach, powered by GraphRAG (Graph Retrieval-Augmented Generation), has shown a remarkable 30% improvement in reasoning capabilities compared to traditional approaches. As Alcaraz emphasized, “Agentic is about using GenAI for running processes” – it’s not just about analysis, but about taking action.
The GraphRAG Revolution
What makes GraphRAG particularly powerful is its ability to combine symbolic reasoning with neural approaches:
- RDF Graphs provide the logical foundation for deductive reasoning and are strategic for data interoperability
- Property Graphs enable flexible data modeling and efficient querying
- Cross-modal integration helps synthesize information across different data types
Memory Architecture and Planning
Building truly intelligent AI agents requires sophisticated memory architectures and planning capabilities. Plan-on-Graph represents a recent breakthrough in how AI agents approach complex tasks, enabling:
- Decomposition of complex queries into manageable sub-tasks
- Iterative exploration of reasoning paths
- Dynamic adaptation to changing contexts
- Consistency maintenance through knowledge graph integration
Cross-Modal Integration
Modern AI agents must process and synthesize information across multiple modalities:
- Text and structured data (schedules, regulations)
- Real-time sensor data (train locations, platform occupancy)
- Visual information (station maps, maintenance images)
- User interaction histories
The power of this integrated approach lies in how these components work together. Strategic intelligence guides overall decision-making, tactical execution ensures effective implementation, and interface intelligence maintains meaningful user interactions.
This comprehensive framework, supported by GraphRAG technology, enables AI agents to not just process information but to take meaningful action based on their understanding.
Risk Management and Control Framework
AI agent implementations face several critical risks that need careful management. Technical risks include data quality issues, with entity resolution accuracy often only reaching 60%, and scaling challenges in graph traversal for large knowledge bases. Business risks center around potential “Mega AI black holes” absorbing organizational knowledge without proper controls, and operational dependencies on external AI systems.
To mitigate these risks, organizations should: build strong ontological cores as knowledge foundations, implement comprehensive Entity-Resolved Knowledge Graphs (ERKG) with multiple validation layers, establish clear semantic boundaries for data access, and deploy robust governance frameworks.
Additionally, organizations should use standardized specifications like DPROD (Data Product Ontology) for consistent data management and maintain clear separation between agent autonomy and human oversight.
Implementation Challenges and Solutions
The implementation of these architectures presents several key challenges:
Scale and Performance:
- Efficient graph traversal for large-scale knowledge bases
- Real-time response requirements for user interactions
- Resource optimization for memory-intensive operations
Consistency Management:
- Handling conflicting information
- Maintaining temporal consistency
- Resolving ambiguities in cross-modal data
Integration Complexity:
- Connecting diverse data sources
- Building coherent data products
- Managing real-time updates
- Ensuring system resilience
To address these challenges, Alcaraz recommends a layered approach that separates strategic intelligence (long-term planning and learning) from tactical execution (immediate response and action). This separation allows for more efficient resource allocation while maintaining system coherence.
Conclusions and future directions
The dawn of the AI agent economy marks a pivotal moment in the evolution of artificial intelligence. As we move beyond the era of unlimited training data, success will increasingly depend on how well organizations can structure and leverage their existing knowledge. The TrenIA experiment demonstrates that by focusing on the ontological core and embracing agentic systems, organizations can transform their service delivery capabilities.
Looking ahead, several key factors will determine success in this new landscape:
- The ability to identify and structure domain-specific knowledge
- The development of sophisticated reasoning and planning capabilities
- The integration of multiple data modalities and knowledge sources
- The maintenance of consistent and explainable AI systems
Organizations that embrace these principles and invest in their ontological core will be best positioned to thrive in an economy where AI agents become integral to service delivery and business operations.
Ready to Navigate the AI Agent Economy?
The shift towards reasoning AI agents and the power of the semantic layer are transforming industries. If you’re an organization leader looking to unlock the potential of your data and build a robust foundation for AI innovation, we invite you to connect with us. Learn how WordLift can help you define and implement your ontological core, build a powerful semantic layer, and thrive in the emerging AI Agent Economy.
Contact us today to explore how we can partner on your journey.
