The last two years in artificial intelligence adoption have been all about generative AI, now, another technological wave – AI agents – promises to disrupt the tech landscape, potentially making ‘agentic’ the defining term of 2025.
The rise of AI agents
AI agents are software systems capable of performing tasks autonomously, leveraging the capabilities of generative AI and foundation models. Unlike static tools, AI agents possess a degree of ‘agency’ – the ability to take independent action based on goals or inputs, often without requiring constant human intervention.
One definition (Kapoor et al, 2004) describes AI systems as ‘agentic’ when they have three characteristics; they can autonomously tackle challenging objectives in complex settings without explicit guidance; they can follow natural language instructions and operate independently; they can utilise tools, like web search or programming, or demonstrate planning capabilities. Agents can generally be divided into two main types: software agents and embodied agents. Software agents operate on computers or mobile devices, leveraging apps to perform tasks such as organising travel or automating office work like sending emails and managing sequential activities. Embodied agents exist within 3D environments, such as video games or robotic systems.
One approach for agents involves making them multimodal, enabling them to handle language, audio, and video seamlessly. For instance, in Google’s Astra demonstration, users could use a smartphone camera to capture objects and inquire about them. The agent was capable of responding to inputs in text, audio, or video formats.
The potential of these agents ranges from mundane task automation to profound changes in how businesses and individuals operate. While consumer-facing AI agents aim to simplify everyday life, B2B agents are poised to revolutionise business processes by functioning as autonomous operators. Sarah Friar, OpenAI’s chief financial officer, believes such AI agents will shape the future. She envisions agents serving as personal assistants, researchers, or even task managers.
“In 2025, we will see the first very successful agents deployed that help people in their day to day,”
Friar told the Financial Times.
But what exactly might these agents look like, and what could they achieve?
Real-world examples of AI agents today
AI agents are already beginning to make their presence felt. Start-ups and tech giants are leveraging their potential across industries. For instance:
Recruitment and onboarding: AI agents are being used to identify, engage, and even interview job candidates. Companies like HireVue employ AI to streamline hiring, saving recruiters time on repetitive tasks.
Software development: AI agents, such as Google’s Jules, assist in software development by generating code, debugging, and providing real-time solutions based on user queries.
Customer support: Tools like Zendesk AI use agents to handle routine customer queries, escalate complex cases, and even predict customer sentiment during interactions.
Gaming: AI agent MineDojo was developed for Minecraft. Trained with extensive data from online sources, the agent acquired various skills and could perform tasks such as building structures to pen llamas or managing lava collection with a bucket.
These applications illustrate a fundamental shift with artificial agents not just envisaged as augmenting humans but beginning to replace them in specific tasks.
Consumer-facing agents: assistants for all?
While business applications have dominated the early narrative around agents, consumer-facing AI agents are on the horizon. Imagine an AI assistant capable of managing your schedule, booking travel, and even handling personal finances – all through natural language conversations. Companies like OpenAI, Google, Amazon, and Meta are vying to create general-purpose agents to serve as everyday helpers. For example, Google’s Duet AI integrates with Google Workspace, assisting with tasks like summarising emails or creating presentations. Similarly, Alexa and Siri are evolving from voice-command tools into more sophisticated, proactive agents.
These developments could potentially make personal AI agents redefine productivity and convenience for consumers.
Challenges to adoption
Despite their promise, the widespread adoption of AI agents faces significant hurdles. The prospect of replacing entire organisational teams with autonomous agents is unlikely to be welcomed by managers, not just from job security fears but also from a lack of understanding about how to integrate AI into existing workflows. CEOs may push for adoption in their corporate structures, but more likely, start-ups and new organisational structures will emerge to harness AI agents effectively.
There are also questions about trust and control, particularly in systems where agents interact with one another without human oversight. As agents take on more responsibilities, how do we ensure they act ethically and reliably? A ‘multi-agent ecosystem’ could lead to scenarios where improperly trained agents make costly or dangerous decisions. Stanford professor Silvio Savarese highlights the importance of guardrails.
“We need systems that behave appropriately,”
he warns, advocating for designs that require agents to flag uncertainties and seek human guidance. Without such safeguards, the potential risks and harm from AI agents escalates.
Future-gazing: the agentic revolution
Looking ahead, the agentic revolution could redefine industries, organisational structures, and daily life. Here’s what the future might hold:
Zero-employee companies: Some start-ups envision entirely autonomous organisations, run by agents that handle everything from product design to customer support with the role of CEO shifting to being someone who knows how to direct AI systems. While these “ghost companies” may save costs, they also raise existential questions about the role of humans in the economy.
Multi-agent ecosystems: As agents become more specialised, they could increasingly collaborate with one another. For instance, a supply chain agent might interact with a customer support agent to resolve delivery issues without human input. Research into controlling these systems will be critical to prevent unintended consequences.
Hyper-personalised services: Consumer-facing support agents could learn individual preferences to deliver highly tailored experiences. From personalised travel itineraries to curated healthcare recommendations, agents could transform the way we interact with technology.
Redefining human work: With agents taking over routine tasks, humans may focus on creative, strategic, and interpersonal work. This shift could lead to new roles centred around supervising, auditing, and enhancing AI systems.
Changing regulatory frameworks: Governments and organisations will need to establish policies to govern the ethical use of AI agents, addressing issues like bias, accountability and, critically, job displacement.
The rise of artificial agents could be heralding a new era of automation and innovation. In 2025, ‘agentic’ may indeed become the defining word of the year, a forecast of a world where autonomous agents reshape how businesses and individuals operate.