Agentic systems will replace some roles. However, they will also create new opportunities to manage and develop the intelligent systems and ensure they remain on track.
As artificial intelligence advances, we are moving from applications focusing on content generation to actionable knowledge workers. Generative AI tools like ChatGPT have become integral to work, with 28% of employees in the U.S. already utilizing these technologies. The next phase poised to reshape the future of work is agentic AI. Gartner describes the latter as systems that autonomously plan and take actions to meet user-defined goals. By 2027, Deloitte predicts that 50% of companies currently utilizing generative AI will adopt agentic AI.
Anatomy of Agentic AI
Agentic systems are distinct as they are purpose-built to facilitate the enterprise adoption of AI technologies. The agents are adaptable and are able to respond to changing business conditions, such as automatically navigating new system functionality and capabilities. A common misconception of agentic AI is that it only combines agent-oriented architecture with centralized coordination to achieve business process automation goals. However, the recent introduction of Model Context Protocols (MCP), which standardize how AI applications connect with other systems, enables agentic agents to operate independently. This has raised concerns about the risks from agents being able to interact with any system without human intervention.
See also: Beware of Vengeful AI Agents
Keeping Agentic AI on track
Agentic AI has vast potential in helping drive productivity and, unlike individuals, it doesn’t require vacation days. However, before these systems can be deployed, extensive continuous testing, observability, and monitoring are required to validate that they perform exactly as expected. Machine workers, like employees, require ongoing evaluation, supervision, and regular training to ensure they meet the needs of the business. By focusing on the areas outlined below, organizations can prepare for this next wave of intelligence.
Data Quality and Governance: Data quality determines how the system learns and acts. Therefore, before integrating agentic AI, organizations must audit data sources and implement a quality assurance process, along with guardrails, to prevent it from doing the wrong thing.
Ethics and Trust: In order for agentic AI to make sound judgments, it’s important that the model data is free of bias. Addressing this requires continuous monitoring and establishing a code of conduct. Additionally, enterprises need to delineate the actions that result in the system being immediately terminated. For example, preventing the automatic transfer of large sums of money exceeding a set amount or to accounts in certain countries.
Privacy & Security: As the AI agent interacts with multiple systems, this is yet another threat vector for cybercriminals to exploit. If bad actors gain access, they can steal data or take control of the machine. Implementing robust cybersecurity guardrails is vital, as undoubtedly, bad actors will exploit agentic capabilities to aid their nefarious efforts.
Decision Making: People need to understand exactly how the agentic system makes decisions. Explainable AI is critical for building trust and mitigating concerns around the machine going rogue. This must be resolved before regulated industries, like healthcare and financial services, can deploy agentic systems into their workflows.
Preparing for Agentic AI
Ensuring agentic systems are reliable and robust will determine the speed of adoption. Organizations must thoroughly test the autonomous, adaptive, and decision-making nature of these systems before integrating them. Once in place, continuous monitoring is mandatory.
Initially, agentic capabilities will be integrated to automate administrative workflows. Below are some potential use cases you can expect to come to fruition.
- Software Development: A coder can enter ideas for software through a prompt, and then the agentic engineer creates code. This automates several steps of the software development lifecycle and can be used to find and fix bugs or optimize legacy systems.
- Predictive Maintenance: Agentic models will orchestrate data from various sources to obtain a detailed analysis of equipment health and then proactively schedule maintenance before any failure, helping extend the life of the manufacturing plant.
- Logistics: Warehouse and supply chains will continue to be at the forefront of advancing automation to create more responsive systems. Agentic AI will manage warehouse operations by analyzing data from sources, including traffic patterns, weather, and order volumes, to make better decisions. Imagine a warehouse manager struggling with disruptions due to the impact of a hurricane. The agentic system could analyze data to reroute deliveries, saving time and reducing costs.
The potential of these systems is vast. However, the increased complexity and autonomy create new challenges and risks. Before enterprises can harness the full capabilities, they must conduct extensive testing to ensure the technology performs exactly as expected and that it solves issues, rather than creating them. Continuous testing is essential for all AI augmented systems.
Man and Machine
Agentic AI will introduce a new wave of innovations. From a human perspective, the technology will automate tedious administrative tasks, freeing employees to focus on more strategic or creative endeavors. Undoubtedly, agentic systems will replace some roles. However, they will also create new opportunities to manage and develop the intelligent systems and ensure they remain on track. Progress will be similar to autonomous driving; incremental steps towards ever-increasing levels of autonomy. By the 2030s, the workforce will be a blend of human and machine-driven skills, with people firmly in charge.