Chapter 10: The Creation of AGI from Agentic AI

Explore the evolution from agentic AI to the possibility of Artificial General Intelligence (AGI)—capable of human-like understanding, reasoning, and adaptation across diverse contexts. 

Transcript

SPEAKER_00 0:00

Chapter 10 The Creation of AGI from Agentic AI As we have explored in this book, the story of AI is a story about aspiration and innovation, the desire to create systems that can compute and comprehend, respond and reason. Agentic AI shows us a glimpse of what is possible and moves us towards the frontier models that pave the way to artificial general intelligence, AGI. AGI will extend EAI's trajectory towards systems capable of human-like generalization, learning across domains, forming abstract concepts, and acting with autonomy. Enterprises looking to deploy intelligent systems at scale must plan for the governance, orchestration, and life cycle management of agent networks, not just individual models. As these frameworks evolve, they serve as building blocks on the path from agentic AI toward broader, more adaptive cognitive systems, systems that will challenge traditional architectural operational and governance models. This chapter explores the shift from agentic AI, systems that pursue goals and learn from human guidance, toward the promise of AGI, a form of intelligence capable of human-like understanding, reasoning, and adaptation across diverse contexts. It highlights the technical and ethical foundations required for this transition and explores the ongoing debate about whether AGI will arise from scaling up current models or from breakthroughs in new architectures, reasoning abilities, and emotional intelligence. The path from narrow AI to AGI is represented visually on page 165. We start in 1980s to 2000 with rule-based AI, which includes fixed logic systems and deterministic decision making. Next in the 2010s is machine learning, with predictive systems trained on structured data. In 2020s, we see generative and agentic AI with autonomous systems capable of planning and goal-directed reasoning. Finally, in the future, we have artificial general intelligence, which includes systems with flexible cross-domain learning, self-improvement, and moral reasoning. Agentic AI, the bridge towards AGI. While agentic AI can perform task-specific actions, it has domain constraints and cannot learn across domains. Its reasoning is contextual but not conceptual. It cannot abstract principles or generalize learning. In contrast, AGI can learn, reason, and adapt across a broad range of tasks. AGI approaches the flexible intelligence characteristic of human cognition. The transition from agentic AI to AGI is not simply a matter of more data, but of architectural approach. Researchers debate whether AGI will emerge from scaling current foundation models or require a new method of reasoning, experience, and emotional intelligence. Two different approaches are dominating the early thinking on this topic. One, scaling hypothesis predicts that AGI will emerge from continued scaling of today's large language and multimodal models without requiring new algorithms. In scaling laws for neural language models, the authors explain the relationship between model performance and three factors. Model size, dataset size, and compute capacity. Our results strongly suggest that larger models will continue to perform better and will also be much more sample efficient than has been previously appreciated. Big models may be more important than big data. In this context, further investigation into model parallelism is warranted. Deep models can be trained using pipelining, which splits parameters depthwise between devices, but eventually requires increased batch sizes as more devices are used. 2. Discontinuity Hypothesis predicts that AGI will not emerge simply by scaling up existing LLMs or neural architectures, but will require fundamentally new paradigms, architectures, or forms of cognition. Some experts argue that scaling neural networks will not deliver AGI because they lack structured reasoning, causal models, and compositional generalization. Essential features of human cognition. Gary Marcus, for example, has asserted the importance of symbols over neural networks in furthering AI. Symbols, computer internal encodings, like strings of binary bits that stand for complex ideas, still far outstrip current neural networks in many fundamental aspects of computation. They are much better positioned to reason their way through complex scenarios, can do basic operations like arithmetic more systematically and reliably, and are better able to precisely represent relationships between parts and whole. Essential both in the interpretation of the 3D world and the comprehension of human language. They are more robust and flexible in their capacity to represent and query large-scale databases. Symbols are also more conducive to formal verification techniques, which are critical for some aspects of safety and ubiquitous in the design of modern microprocessors. To abandon these virtues rather than leveraging them into some sort of hybrid architecture would make little sense. In all likelihood, the path from agentic AI to AGI will not be a simple one, but instead a mix of model and capability expansion, driven by improvements in data quality, model interpretability, the availability of compute, and leveraging symbolic reasoning beyond traditional neural networks. The role of agentic AI and enterprise orchestration. As agentic AI deployments mature, they provide a first step in enabling hybrid approaches toward the development of AGI. With agentic AI, complex human-level tasks are naturally decomposable and can be distributed. Specialized agents can be used for functions such as planning, researching, coding, verifying, simulating, and managing. The agents perform discrete functions and can operate in parallel and collaborate, reducing the cognitive and computational burden on any single model. In parallel, specialized agents can be improved independently and reused across tasks. Agents can also adopt distinct learning approaches, leveraging reinforcement learning for optimization, symbolic reasoning for logic, or unsupervised learning for discovery. This modularity enables the creation of a hybrid learning system. In combination with the agents, the orchestration layer provides a central coordination mechanism. The orchestrator handles task management by decomposing and assigning tasks, scheduling, and allocating resources. It also manages communications by routing contacts between agents, and provides oversight for validating outputs, monitoring performance, and managing the system lifecycle. Crucially, the orchestration layer also enables learning orchestration, enabling the entire agentic system to improve over time based on distributed experiences. Yet, it is essential to recognize that orchestration alone does not equate to cognition. Many current systems lack long-term planning, persistent memory, reasoning, and model-based causal understanding. From a design perspective, the orchestrator should evolve from a simple task scheduler into a metacontroller with core capabilities such as goal management, dynamic resource and role allocation across agents, reflective error and feedback loops, and continuous evaluation assurance pipelines. This modular orchestrated approach forms a viable foundation for training hybrid AGI systems. In the following feature, find out how Portuguese municipality has improved productivity and cut operational costs using AI-driven automation. Case study a Portuguese municipality. As part of the urban agglomeration of Greater Lisbon, the municipality employs a workforce of over 1,000 employees. For the public sector organization, managing the growing complexity of digital operations was problematic. Fragmented systems and manual workflows made it difficult to classify and prioritize requests, track interventions, or maintain accurate records. Data existed, but insight did not. Without integrated oversight, teams were forced into reactive decision making and struggled to maintain compliance across evolving regulatory standards. The result was a lack of transparency and accountability, an obstacle that modern AI governance models are uniquely designed to overcome. To address this, the organization established a unified AI-driven management framework capable of orchestrating workflows, classifying requests, and predicting service demand through pattern recognition. By introducing structured governance and intelligent automation, it gained real-time visibility into performance, resource allocation, and compliance adherence. Instead of manually managing processes, the enterprise now operates through a self-optimizing system that learns from each interaction. Agentic AI components continuously analyze performance data, refine workflows, and flag emerging inefficiencies, paving the way for adaptive service management aligned with enterprise strategy. These developments represent more than operational improvement. They signal a step toward AGI-like enterprise intelligence. With AI now embedded across service management, the system not only responds to user requests but anticipates them, analyzing context, predicting outcomes, and coordinating across teams. As these agentic systems evolve, they form the foundation of an organization capable of learning, reasoning, and improving at scale. What began as an effort to streamline IT operations has become a model for how intelligent governance, human oversight, and adaptive AI can coexist, creating an enterprise where automation doesn't replace people but empowers them to think and act with greater intelligence. By the numbers, the impact of intelligent governance. 60% faster request resolution after replacing manual workflows with AI-driven classification and prioritization. Up to 40% reduction in operational costs through automation of contract, asset, and service management. 100% visibility into performance metrics and compliance adherence across all service domains. 50% fewer process eras due to automated validation, audit trails, and real-time anomaly detection. Continuous learning loop established. Agentic AI systems now refine workflows autonomously based on data trends and feedback. Cross-functional collaboration improved across departments, replacing silos with transparent adaptive workflows. Data as an enabler to AGI, fueling cognitive scale. The foundation of the evolution to AGI remains data, where data quality, diversity, and governance are the primary enablers. Agentic AI thrives on structured and semi-structured data within defined operational boundaries. AGI, however, requires a richer and more representative data set capable of supporting higher order reasoning. This is why the need for data is only growing in importance as technology evolves. Data for AGI must capture context, causality, and ethics. This requires data frameworks, federated data sharing models, and sovereign data infrastructure that ensures responsible access and use. The OECD AI Principles and ISO slash IEC 42001 2023 standard both emphasize that AI systems should operate under well-defined data governance mechanisms, ensuring fairness, accountability, and traceability. Privacy-enhancing technologies also play a critical role. As AI moves closer to general cognition, maintaining data ethics and integrity becomes as essential as performance metrics. Governance and ethics. Aligning autonomy with accountability. As agentic AI systems become more independent, questions of control, accountability, and oversight are key, as we reviewed in chapter four. However, the challenge increases as we approach AGI, which has a level of autonomy beyond human capability. Ethical frameworks for enterprise AI have historically focused on fairness, transparency, and explainability. However, as systems begin to make complex decisions, it will be important to ensure that AI systems goals remain consistent with human values. As discussed in Chapter 6, international governance efforts are converging around this challenge. The EU Artificial Intelligence Act, 2024, establishes tiered risk categories for AI and mandates strict oversight for high-risk applications. The UNESCO Recommendation on the Ethics of Artificial Intelligence, 2021, calls for human rights-based governance, while NIST AIRMF, 2023, introduces trustworthiness as a measurable element. These initiatives ensure that AI systems remain subject to human intent and oversight even as their capabilities expand. From a policy perspective, AGI could enable a new level of autonomy, in which systems might adapt or self-improve in unpredictable ways. Anticipating this, researchers in AI safety are looking into new standards. Again, this needs to happen in parallel with the technology evolution so that, once AGI is ready, the standards are prepared as well. The enterprise relevance of AGI-like systems. As enterprises deploy increasingly autonomous and interconnected agentic AI systems, many are already encountering early signs of artificial general intelligence in practice, though not in name. What we call AGI-like systems are emerging organically across organizations, intelligent agents that learn from multiple data sources, coordinate decisions across departments, and adapt their behavior based on shifting objectives or external context. These aren't isolated tools, they're self-optimizing networks of intelligence, continuously refining how work gets done. The shift is subtle but profound. Where once automation replaced narrow, repetitive tasks, Agentic AI now integrates reasoning, planning, and self-correction. A financial services team might deploy agents that analyze customer sentiment, predict churn, and autonomously generate retention strategies, activities that span marketing, risk, and compliance in a unified feedback loop. In manufacturing, AI systems can already interpret sensor data, reallocate supply chain resources, and flag ethical sourcing risk before human teams intervene. The line between specialized automation and enterprise scale cognition is blurring rapidly. This evolution carries immense strategic opportunity, but also risk. Without effective governance, enterprises may find themselves managing systems that learn and act beyond their intended design. As noted by Gartner, over 80% of organizations pursuing AI at scale cite governance and transparency as their biggest obstacles to adoption. Managing AGI-like systems requires new models of accountability and oversight. Traditional IT management was built for static systems. Modern intelligence ecosystems demand adaptive governance, where oversight mechanisms evolve alongside the AI itself. Explainability, auditability, and feedback loops must become design features, not afterthoughts. And as these systems begin to reason across domains, human judgment must remain in the loop, not to slow decisions, but to steer them. In this sense, AGI is not a distant horizon, but a growing enterprise reality. The organizations that will lead in this era are those that recognize intelligence as infrastructure, something to be governed, integrated, and continuously improved, just like data or cybersecurity. The result is not machines that replace human capability, but intelligent systems that amplify it, scaling enterprise insight, accelerating transformation, and building a foundation of trust for whatever comes next. In the feature below, a top South African university is using an AI-powered automated service desk to transform student experiences and ensure business continuity when crisis hits. Case study a South African University. The value of machine learning is phenomenal in our student community, as evidenced by the wide use of our virtual agents and knowledge articles. Without machine learning and AI, there is absolutely no way we could support our end users with the few dedicated agents we have. Change and Configuration Manager, University in South Africa. One of Africa's top universities produces research to find solutions for pressing issues. The university teaches in the classroom, online, and embedded in communities. Faced with unprecedented skill and digital inequality, the university set out to transform how it delivered education and support in an increasingly hybrid world. With a student population exceeding 130,000 and only a handful of dedicated support agents, manual processes were no longer sustainable. Limited access to devices and connectivity deepened the digital divide, while non-technical departments struggled to adapt to remote workflows. The challenge wasn't simply technological, it was structural. The institution needed an intelligent operating model capable of supporting both academic and administrative continuity while empowering every student to participate fully in a connected ecosystem. To meet this challenge, the university re-engineered its digital service framework around automation, governance, and intelligence. AI-driven systems were introduced to classify requests, predict demand, and route tasks automatically, allowing a small team to manage massive volumes of support interactions in real time. When COVID hit, machine learning capabilities accelerated adaptation during the global shift to remote learning, automating VPN access, provisioning laptops, and optimizing connectivity support for thousands of students. Agentic AI became the invisible backbone of the university's digital infrastructure, orchestrating workflows across departments and extending intelligence to non-IT functions such as enrollment, student services, and library operations. Each system learned from interactions, improving accuracy, responsiveness, and fairness across the institution. What began as a crisis response evolved into a model for AGI ready education, one where adaptive intelligence enhances both scale and equity. Today, service request volumes have increased dramatically, yet efficiency and transparency have improved in equal measure. AI not only supports staff and students, it collaborates with them, learning from patterns, context, and feedback to anticipate needs and streamline decision making. The university's transformation demonstrates how intelligent governance and agentic automation can bridge human and digital capability, laying the foundation for an academic ecosystem where intelligence is distributed, collaborative, and continually self-improving. Defining the future of AGI Beyond the Technical Horizon. The pursuit of AGI is as much a philosophical and social journey as it is a technical one. While some view it as the logical outcome of scaling current architectures, scaling hypotheses, others see it as a redefinition of intelligence itself, a step towards systems that possess intentionality, moral reasoning, and self-awareness. Discontinuity Hypothesis. If agentic AI represents the automation of action, AGI represents the automation of understanding. Future AI systems must not only think and learn, but also align with collective human values, which is a challenge that will define the next decade of enterprise AI policy and innovation. The transition from agentic AI to AGI is not solely about machines surpassing human capability, it is about how human and machine intelligence evolve together. As AI systems become more capable, they will also reshape the roles of the human workforce. And for this reason, human oversight must remain a priority. From a policy and governance perspective, as well as an execution and operations perspective, roles that are commonplace today may no longer be required, but equally new roles will emerge. These could range from training and managing the agentic AI capabilities to ensuring data quality and governance to leveraging AI to create new products and capabilities that are not envisaged today. With each shift in technology comes new opportunities, and AGI will be no different. The future of work will not be defined by human replacement, but by human machine partnership. Designing effective collaboration models between human specialists and generalist AI systems means clearing defining roles, authority, and accountability within shared workflows. It also means investing in continuous training and digital fluency so teams understand how to question, interpret, and guide AI outcomes responsibly. In a world where AI can learn faster than its creators, sustaining trust and ethical oversight becomes the anchor, ensuring that intelligence serves humanity's goals, not the other way around. In this chapter, we explored the evolution from agentic AI to the possibility of AGI, which would exhibit human like reasoning, learning, and adaptability across diverse contexts. This transition will involve not just scaling data, but also integrating improved data quality, model interpretability, and symbolic reasoning, paving the way for hybrid EAI systems that can tackle complexity. Task through specialization and collaboration. Looking ahead and considering that AGI is not the replacement of human intelligence, but its next great amplifier, data that is managed and governed in EIM will play a crucial role as an enabler of AGI. It will provide the foundational knowledge necessary for these advanced systems to learn, make sound decisions, and adapt in increasingly complex and dynamic environments. Investments made today in agentic AI will be valuable in the evolution to AGI. In the following feature, find out how a Mexican retail analytics provider transforms sales data into actionable insights to boost revenues. Case study A Mexican Retail Analytics Company. Using AI, we have achieved a quantum leap in performance and dramatically reduced query response times. Our clients now have the critical data they need at their fingertips to optimize their sales revenues. Manager Director, Retail Analytics Company. The company delivers an AI-powered analytics platform that enables retailers to optimize sales performance and decision making. Serving over 130 leading consumer brands across Latin America, the cloud-based system unifies sales and inventory data, analyzes purchasing behavior in real time, and generates predictive insights to guide smarter, faster operational decisions. The company set out to solve one of Commerce's most persistent challenges, balancing supply and demand with precision. In a world where consumer preferences shift by the minute, the ability to synchronize sales, inventory, and pricing decisions is critical. Yet, legacy systems built on traditional databases struggled to scale with the volume and velocity of data being generated. As transaction loads soared, the company found its architecture strained, queries slowed, insights lagged, and the ability to make real-time adjustments faded. For businesses depending on timely intelligence, this was more than a technical bottleneck. It was an existential risk to competitiveness. To move beyond reactive reporting, the company reimagined its analytics platform through the lens of AI. Machine learning and adaptive algorithms now process billions of records daily, identifying emerging demand patterns and optimizing distribution at scale. The introduction of AI-powered simulation tools enabled predictive pricing strategies that once took hours to calculate to be executed in seconds. These systems learn continuously from historical data, testing new variables, and refining elasticity models across thousands of products and regions. The result is a platform that doesn't just describe what happened, it anticipates what will happen next, transforming static analytics into a living, learning system of intelligence. This evolution marks a shift from analytics to cognition, a step toward enterprise-level AGI. By embedding reasoning, prediction, and self-optimization into its operations, the company has built a digital nervous system capable of adjusting in real time to market behavior. Every new transaction becomes a learning signal, strengthening the feedback loop that guides future strategy. With each iteration, the system grows more attuned to human decision making, amplifying, not replacing it. What began as a search for faster insights has evolved into a glimpse of the cognitive enterprise, one where intelligence is distributed, collaborative, and continuously self-improving. The Fast Five Download. One, Agentic AI as an enterprise foundation. Agentic AI serves as a crucial stepping stone towards AGI by enabling task decomposition, specialization, and collaboration among autonomous agents. This modular approach lays the groundwork for more flexible, scalable, and human-like intelligence systems. 2. Competing paths to AGI. The transition to AGI is shaped by two dominant hypotheses. The scaling hypothesis, AGI emerges from scaling current models, and the discontinuity hypothesis, AGI requires fundamentally new architectures and reasoning. Executives should monitor both trajectories for strategic planning and investment. Three, data strategy is critical. Progress toward AGI will be driven by data quality, diversity, and governance. Organizations must prioritize robust data frameworks, privacy-enhancing technologies, and compliance with evolving international standards to fuel advanced AI capabilities. Four, governance and ethics at the core. As AI systems become more autonomous, aligning their actions with human values is essential. Executives must champion strong governance, ethical oversight, and risk management frameworks to anticipate regulatory requirements and societal expectations. 5. Prepare for workforce and policy shifts. The evolution from agentic AI to AGI will redefine workforce roles and policy landscapes. Proactive investment in talent, change management, and ongoing human oversight will be vital to harness AGI's potential while mitigating risks.