Path from Agentic AI to Autonomous AI

Building current Agentic AI capabilities to define the future of autonomous AI

The ERA of AI: A New Dawn

The dawn of the AI era is upon us, a transformative epoch that is reshaping industries, societies, and individual lives. The proliferation of AI technologies, from machine learning algorithms to sophisticated neural networks, has fundamentally altered our interaction with technology. This era is characterized by the rapid integration of AI into everyday applications, demonstrating its potential to enhance productivity, drive innovation, solve complex problems and push the boundaries of science and research.

If you are impressed with the very first paragraph of this paper, then this is an example of one of the many aspects of how AI is making us better individuals, certainly making myself – a non native English speaker- sound articulate and eloquent and helping me share my thoughts about how Agentic AI is evolving to become the technology equivalent of “Art mimics life” and how Autonomous AI is set to look like, not in the very far future.

The purpose of this paper is to outline agentic design patterns and roles relevant to create agentic structures and networks supporting value creation and delivery at small and large scale and to set a baseline for best practices to build agentic to autonomous systems vision.

What is Agentic AI?

Agentic AI represents a class of AI systems designed to achieve specific goals with limited direct supervision by adapting to complex environments. These agents are characterized by their ability to perceive their surroundings, make decisions, and take actions autonomously. Unlike traditional AI, which follows predefined rules, agentic AI systems exhibit a higher degree of flexibility and adaptability, making them suitable for dynamic, real-world applications.

Agentic AI functions through a sophisticated interplay of perception, decision-making, and action. These systems employ advanced sensors (picture, video streams, voice…) and data collection tools to perceive their environment accurately. With the aid of machine learning algorithms and neural networks, they process this data to identify patterns, make informed decisions, and predict outcomes. The capabilities of agentic AI are further enhanced by feedback loops that allow the system to learn from its actions and refine its strategies over time. This continuous learning process enables agentic AI to adapt to new challenges and optimize its performance, making it a powerful tool for navigating dynamic and unpredictable environments.

The importance of Agentic AI cannot be overstated in today's rapidly evolving technological landscape. Agentic AI holds the potential to revolutionize various industries by offering unprecedented levels of efficiency, adaptability, and autonomy. Its ability to learn from its environment and make informed decisions without constant human oversight enables organizations to tackle complex problems with greater agility and precision. From optimizing supply chains to enhancing customer experiences, agentic AI empowers businesses to innovate and thrive in competitive markets. Moreover, by automating routine and mundane tasks, it frees up human resources to focus on higher-value activities that require creativity and critical thinking, ultimately driving economic growth and improving quality of life.

Current examples of Agentic AI include agents that enhance productivity by summarizing and generating content, coordinating and facilitating meetings, and gathering and mining knowledge. This does note only demonstrate AI as a unified Natural Language User Interface within operating systems, edge devices, platforms, and applications, but also a large range of basic capabilities that support the average user on a day to day basis with their tasks.

Other examples are defined by niche business cases relevant to productivity enhancement and process optimisation in specific areas or functions creating AI based capabilities to provide employee self-service, deliver adaptive learning material and many more in Finance, Legal and HR.

Building Agentic AI

Building agentic AI involves several critical steps that are essential to ensure the agent’s mission is clear and set to deliver outcome within the appropriate environment model and domain of knowledge.

·       Goal Definition: Start by defining specific, measurable, achievable, relevant, and time-bound (SMART) objectives. This ensures that the AI agent has a clear understanding of what it needs to accomplish.

·       Data analysis: Analyse historical data to uncover trends and patterns that can inform the goal-setting process. This helps in setting realistic and attainable benchmarks.

·       Purpose-Driven Decision making: Ensure that every decision made by the AI agent aligns with the defined objectives. This involves creating a framework that guides the agent's actions towards achieving the goal.

·       Environment Modelling: Create a virtual representation of the environment in which the agent will operate.

·       Learning Algorithms: Implement machine learning algorithms that enable the agent to learn from interactions within its environment.

·       Adaptability: Equip the AI agent with the ability to adapt its actions based on real-time feedback from the environment. This helps in maintaining alignment with the goal despite changes in conditions.

·       Autonomy: Allow the AI agent to optimize its strategies independently, reducing dependency on human intervention. This involves enabling the agent to take goal-directed actions with minimal oversight.

·       User Experience: Incorporate user feedback and past experiences to refine the AI agent's performance. This helps in continuously improving the agent's strategies and actions.

·       Monitoring and Evaluation: Continuously monitor the AI agent's performance and evaluate its progress towards the goal. This helps in identifying areas for improvement and making necessary adjustments.

·       Testing / tuning: the agent’s capability to achieve set goal(s) will depend on the user’s ability for excellent prompting, testing and tuning will ensure the agent is accurate

For instance, BaristAI is an agentic AI solution tasked with automating customer service in a coffee shop, the goal of the Agent to understand the customer order, process payments and fulfil orders. Agent is able to autonomously map customer orders to pre-sets of items in the menu and can logically map special requests accordingly such as types of milk accordingly to adapt with specific customer needs. The product offer (menu) and existing data from previous purchasing processed by humans are a gold mine of information that allows the Agent to provide the best user experience at speed and at scale.

The Evolution Towards Autonomous AI

Agentic AI is poised to evolve into fully autonomous AI systems that mirror the complexity and structure of human societal and business environments. This evolution will involve several key stages:

Reflecting Business and Societal Structures

Autonomous AI will progressively mirror the hierarchical and collaborative frameworks present in businesses and societies. For instance, AI systems within a corporate environment may comprise diverse agents dedicated to specific functions such as finance, marketing, and operations, akin to a company's organizational structure. These agents will engage in collaboration and communication, exchanging information and coordinating activities to accomplish overarching business objectives. In societal applications, autonomous AI could manage public services, healthcare systems, and environmental monitoring, serving as digital counterparts of societal infrastructures, thereby enabling organizations to coordinate more effectively and address larger and more intricate challenges collectively.

Human like integrability and information processing

Perhaps the biggest challenge to globalise digital systems is the complexity of integration across different systems, even within the same organisational structure, often getting a seaming end to end experience for customer is nearly impossible due to different systems being used to execute segments of business processes, these systems often dictate integration interfaces that are not standardised and used different protocols to exchange information. In the Era of AI, two integrations trends will be predominant:

-          Data Democratisation: this concept allows entities, often within a common domain reflecting governance and responsibilities, to share raw data at lowest available grade that will be used to enrich, create insight or even create AI Agents this of course will require the right control and governance tooling to ensure security and compliance

-          AI interfacing: the ability of two or more AI agents to communicate with each other directly within the context of their set goals allowing agents to exchange information which will be processed back to data that enriches both (or multiple parties, for instance two Agentic Finance controllers from a coffee shop and a retailer automatically reconciliating purchase orders of coffee beans over a chat.

Data democratisation is today essential to unearth the value of the data and endeavour into AI transformation, this will shift as more complex AI systems will be built shifting system integration to natural language (if not new AI languages) models.

Agent Roles Needed for Autonomous AI

To achieve true autonomy, several specific roles must be assigned to agents within an AI ecosystem:

·       Interface Agent: The Interface Agent is the intermediary between human users and AI systems facilitating interactions, understanding asks and needs and collecting feedback. Today the Interface Agent is the most common adoption of Agentic AI given it serves as a UI and deliver value directly to consumers. In less complex structures the Interface Agent can also play the role of an orchestrator or a processor, in the coffee shop scenario, in addition to interfacing with the customer, the  Interface AI would provide inputs to Processors to process the order and could directly process the payment. In large and complex structures the Interface Agent role should limited to understanding and conversing with the customer, relying on other systems to ensure processing and execution of the ask. This type of agent is often trained with large language models reflecting the full capability and mission of a business or an organisation

·       Processor: A Processor is an AI agent that provides Intelligent Process Automation capabilities within a niche business domain. This AI is trained with large language models focused on niche knowledge bases focused on specific disciplines, skills or function. In our example the Processor AI is trained managed the assets that produce the coffee orders, think of it as an AI digital Twin of a smart coffee machine. Other examples could vary from large disciplines such as a HR or Marketing trained AI to specific tasks skills or functions like financial reconciliation or operating machinery. A processor will always require a trigger and/or Orchestration by another system to deliver it’s goal and outcomes. The processor uses a toolkit of skills to execute deterministic task in form of processes such as grinding coffee or creating and printing receipts.

·       Orchestrator: an Orchestrator is an Agent that has deep knowledge of existing resources including other Agents, their roles, and capabilities. The orchestrator autonomously coordinate these resources based on information collected by the Interface Agent to ensure fast and accurate processing as well as triggering other capabilities needed such as control and approvals. The Orchestrator is an important cog in large and complex structures where the Autonomous AI architecture will be dictated by business accountability, data residency and system extensibility, for instance after an Interface Agent confirm a customised pre-order of a new hydrogen car the orchestrator will coordinate the resources needed for payments, financing, production, supply chain and third party Agents needed to fulfil the order.

·       Controller: A Controller is an Agent that is responsible to continuously oversee the compliance of the Autonomous AI system, this includes the overseeing of processing of personal data and ensuring compliance and privacy, overseeing accounting activities and application of industry standards for compliance and risk management purposes the Controller Agent provide the ability validate decisions taken autonomously given the non-deterministic nature of AI. In the context of BaristAI, the controller will oversee the application of the best hygiene standards and the decision made in line of use of products that could induce allergy reactions. In a larger environment, the controller will ensure no sensitive or personal data is shared with a third party autonomous AI system as part of an order fulfilment that requires supply chain coordination.

·       Reporter: The reporter is an agent that provides Tactical and Strategical Business Intelligence to enable more accurate decision making for agents, the entire Autonomous AI System and to the humans governing the system, the reporter is an important role in large structures as it will ensure orchestrator have the right insight to coordinate the resources needed, especially in line with processes that are time consuming and require complex tasks to happen to drive the desire outcome.

·       Predictor: a Predictor is a complex AI agent that leverages complex AI and ML toolkit to deliver advanced analytics and forecasting capabilities, the role of the predictor is to provide insights that enable specific agents or the entire Autonomous AI ecosystem to make decision or introduce changes to adapt with the insights. In the context of our AI coffee shop example, a predictor can provide forecasting on when additional coffee beans need to be ordered to cater to the demand, this information can be used by the orchestrator to trigger a processor to order additional stock.

·       Approver: The Approver is the Agent responsible for validating decisions against governance, control and risks. This agent approves the AI process execution, the data outcome and the Natural Language Experience based on specific guidelines and policies defined to ensure the Autonomous AI system behaves, communicates and manage data in line with regulatory requirements. For instance an approver can approve discount or the refund of a coffee.

·       Monitor: The Monitor agent is an AI that continuously observes the logs, performance, actions, triggers, knowledge base accesses, and inputs and outputs of the Autonomous AI system, the monitor’s goal is to guarantee transparency and observability across the Autonomous AI system. The monitor could be pre-packaged as part of the underlying AI platform powering the AI system providing both Technical and functional monitoring capabilities.

·       Investigator: The Investigator agent provides the capability to understand and identify cause issues in order to address them, an example of when an investigator would be triggered is when a customer receives a wrong coffee order. The investigator mission is to identify what went wrong and helps the Autonomous AI or specific AI agents to address the issue and kick off self – healing.

·       Enforcer: The enforcer works closely with the investigator to enforce changes, self-healing and intervenes when the behaviour of the Autonomous AI system or its components does not meet the defined goals or the expected outcomes and metrics. The action taken by the enforcer is defined by the outcome of the investigator and would require approval

The roles of Agents and patterns in which they collaborate would reflect the ideal structure, network and processes to deliver value in the best customer experience possible. This should reflect across the following

·       The way the system is defined to deliver value to the customer

·       The way the system is defined to convey experience to the customer

·       The way the system is defined to reduce risk to the customer, the system itself and the business

·       The way the system is defined to achieve consistency vs adaptability

·       The way the system is set to be sustainable, self sufficient and scalable

Governance, Controls, and Observability

As AI systems become more autonomous, robust governance frameworks and control mechanisms will be essential to ensure their safe and ethical operation. Key considerations include:

·       Transparency: AI systems should be transparent in their decision-making processes, with clear documentation of algorithms and data sources. A robust logging strategy will allow autonomous AI to self-heal and to continuously improve.

·       Accountability: Establishing clear lines of accountability for the actions and decisions made by AI agents across defined autonomous roles and the inclusion of human in Approval chains, monitoring the monitors and reviewer roles

·       Ethical Standards: Ensuring that AI systems adhere to ethical guidelines, particularly concerning privacy, bias, and fairness, the role of policy and regulatory requirements can not be overstated to ensure the system operate in a reduced-risk environment

·       Continuous Monitoring: Implementing real-time monitoring systems to track the performance and behaviour of AI agents, identifying and mitigating potential issues.

What the Future Holds

As we advance into the future, autonomous AI systems will seamlessly integrate into the fabric of daily life, transforming industries and societies alike. In smart cities, autonomous agents will efficiently manage traffic flows, optimize energy consumption, and ensure public safety through real-time data analysis and predictive algorithms. In healthcare, AI will revolutionize patient care with personalized treatment plans, accurate management of medical records, and assistive roles in complex surgeries, thereby enhancing the precision and effectiveness of medical interventions. Corporations will witness unprecedented efficiency as autonomous AI takes charge of everything from supply chain management to customer service, streamlining operations and improving responsiveness. This evolution will require robust governance frameworks, continuous monitoring, and adherence to ethical standards to ensure the safe and responsible deployment of AI technologies. By fostering a collaborative relationship between human and machine, we can harness the full potential of autonomous AI, driving progress and prosperity in an era where innovation is balanced with societal well-being.

The complexity and impact of what AI can do in the future to the betterment of humanity is vast, as AI systems are being built and matured it is important to engineer these components to reflect the best practices that humanity accumulated over centuries of innovation, creativity, organisation, connection and responsibility, these systems will either reflect how we as a society can be functional or dysfunctional, regardless of how much powerful AI as a technology is, it can be hindered by the quality of data we have, are producing and the design, environment and boundaries we set for each AI system we build today to grow and mature toward an Autonomous AI ecosystem.

Conclusion

In conclusion, the journey towards autonomous AI systems presents both immense opportunities and significant challenges. Effective governance frameworks, comprehensive controls, and rigorous observability will be crucial to harness the power of AI responsibly. Transparency, accountability, ethical standards, and continuous monitoring are key pillars in this endeavour, ensuring that AI systems operate safely and ethically.

As AI technologies evolve, their integration into various facets of daily life will transform industries and societies, contributing to smarter cities, advanced healthcare, and more efficient corporate operations. The success of this integration will depend on our ability to balance innovation with societal well-being, fostering a harmonious relationship between human and machine. By adhering to best practices and principles accumulated through centuries of human progress, we can build AI systems that reflect the potential for both functionality and responsibility, guiding humanity towards a bright and prosperous future.