Hero Image
Esteve Almirall & Xavier Amatriain

The new players: Intelligent agents and the future of business

Innovation & technology 17 September 2026

The rise of intelligent agents marks a new phase in AI, shifting from automation to active decision-making. These systems create a new playing field by reasoning, adapting, and interacting with their environment.

Do Better Team

Artificial intelligence is entering a new phase. While chatbots and generative AI have been at the forefront of discussions, a new generation of intelligent agents is emerging as key players. These agents, designed to operate autonomously across various environments, are transforming commerce, customer service, and logistics. At the corporate level, 51% of companies are actively researching the use of artificial intelligence agents in their processes, while 12% have already deployed them at scale (KPMG, 2025). But what does this shift mean for businesses and society?

What are AI agents? Beyond artificial intelligence models

Xavier Amatriain Vice President of Product (AI and Compute Enablement), discussed a common misconception in the industry at the 4YFN event at MWC Barcelona 2025: the difference between AI models and intelligent agents. “Everyone talks about agents, but nobody really knows what they are,” he remarked.

While AI models generate outputs based on data, AI agents integrate models, tools and external signals to act in the world. They don't just react; they have a form of long-term reasoning that lets them make decisions beyond immediate responses. The real challenge, according to Amatriain, lies in designing these agents with the right access to tools and data so they are truly useful for businesses.

When AI is seen as a new form of professional intelligence, it becomes both transformative and irreversible

How does an AI agent work?

An artificial intelligence agent operates through a continuous four-phase cycle:

  1. Perception: it gathers information from its environment through digital interfaces or data streams.
  2. Reasoning: it processes that information using language models, decision trees or neural networks to build a representation of the current state.
  3. Decision: it evaluates possible actions against a goal or utility function and selects the most promising one.
  4. Action: it carries out real tasks, such as sending emails, searching for information, triggering systems or generating reports, and feeds the result into the next cycle.

This ability to chain actions together is what sets AI agents apart from traditional chatbots, which simply answer a query and stop.

AI as professional intelligence: From automation to reasoning

Omar Puertas, an AI expert and partner at Cuatrecasas, offered a perspective that invites a rethink of the role of artificial intelligence agents in organizations. "AI as an assistant or copilot is a distraction. What's truly transformative is that we are reasoning with technology." Rather than simply speeding up existing processes, AI agents can fundamentally change the way organizations think about their operations.

"If AI is merely used as a productivity tool, it won’t be transformative. But if it is seen as a new form of professional intelligence, it becomes both transformative and irreversible," he added. AI's true value lies in its ability to enhance human thinking. For instance, it can generate multiple counterarguments based on specific criteria, allowing professionals to explore perspectives they may never have considered. 

Consider the legal sector. AI is not just executing tasks but critically assessing arguments, strategies, and reasoning structures. This transformation enables businesses to embed AI in their decision-making, moving beyond efficiency gains to true cognitive augmentation. 

The evolution of AI agents: From banking to strategic sectors

Banking has always been at the forefront of AI adoption, from fraud detection to automated customer interactions. However, as Rosa Martínez highlighted, the rise of generative AI is shifting the paradigm. "AI tools are no longer exclusive to tech teams; they are now accessible to almost anyone." 

One of the biggest impacts is in customer service, where AI can respond to complex queries in a structured and attentive way. Beyond customer interactions, generative AI allows banks to take a more proactive approach, identifying needs before they arise and offering personalized financial solutions.

However, Martínez emphasized that companies must use AI purposefully. It's important to determine where generative AI should be applied and where other technologies might be more appropriate. Additionally, businesses must be adaptable, as AI models evolve rapidly. The key to success, she noted, lies in high-quality data: "Accurate, comprehensive, and extensive data are fundamental. Not all organizations are at the same level when it comes to data readiness."

The trend is spreading across sectors. According to IDC (2025), by 2028, 40% of professionals at Europe's top 500 companies will work alongside AI agents to generate reports and manage data. The agentic AI market has already reached $7.38 billion and is growing at 44.8% a year.

Intelligent agent examples: real-world use cases

These are some of the clearest intelligent agent examples in business for 2025-2026:

  • Finance: agents that monitor portfolios in real time, detect fraud and generate financial reports autonomously.
  • Legal: systems that analyze contracts, assess regulatory risks and suggest litigation strategies.
  • Logistics: agents that optimize delivery routes, adapting to disruptions in real time.
  • Customer service: assistants that resolve complex queries and hand off to human agents only when necessary.
  • Healthcare: agents that cross-reference medical records with scientific literature to support differential diagnoses.

Who is accountable when an AI agent fails? Transparency and evaluation

Álvarez-Napagao provided insights into the forefront of AI development. While we now have generative models for nearly everything, the defining feature of intelligent agents is their ability to perceive and interact with their environment in specific ways. Álvarez-Napagao, who has researched intelligent agents for years at the Barcelona Supercomputing Center, highlighted a major inconvenience: “We still don’t fully understand why AI models work the way they do or why they exhibit certain properties.” 

We must think of AI agents as augmenting what we do, but there must always be a human accountable for their decisions

This brings us to one of the most pressing concerns surrounding intelligent agents: accountability. As AI systems gain greater autonomy, the question arises: who is responsible for their actions? Amatriain cautioned that AI does not possess consciousness—nor moral agency. "We must think of them as augmenting what we do, but there must always be a human accountable for their decisions," he said.  

We can consider a hypothetical example of a company training an AI agent to act as a CEO. If such an agent were to make a harmful decision, the issue of legal and ethical responsibility would become highly complex. The same dilemma extends to many other areas where AI-driven decisions are already a reality. What questions arise when AI determines who gets a loan or social welfare? Who gets hired or fired? Or, in even more troubling domains, what happens when a military AI decides what or who is a legitimate target? 

AI that trains itself: The opportunities and risks of self-improvement

Looking ahead, Álvarez-Napagao suggested that AI agents will soon become deeply embedded in everyday life. “Ten years ago, we couldn’t have imagined the level of AI-generated images we have today.” In his view, within another decade, we may even see household, hospital and care robots capable of planning, moving autonomously, and assisting with daily tasks. However, he stressed the importance of user education: if we don’t properly prepare people for how these agents will interact with their environment, “we risk making significant mistakes.” 

One of the most intriguing topics of the discussion was AI's ability to train itself. Amatriain confirmed that most data used to train AI models is already being labeled by other AI models. "This is evolving rapidly. It's exciting to see that these models are on track to discover solutions to problems we haven't even considered yet," he noted. They could, for instance, play a key role in curing diseases by identifying patterns humans would overlook.

This is a brave new world, and it’s happening now

However, this rapid self-improvement also comes with risks. We need to maintain oversight and understand what's happening inside these models. The idea of an unchecked self-improvement loop is concerning, Amatriain noted. While some speculate that certain models have already entered this phase, he emphasized the importance of maintaining human supervision to ensure AI remains aligned with societal needs. 

New trends in the evolution of AI agents

The evolution of AI agents is currently defined by three major trends:

  • Multimodal agents: able to process text, voice, images and structured data within a single workflow, which extends their use to physical and digital environments at the same time.
  • Collaborative agent networks (multi-agent systems): several intelligent agents coordinating with one another to handle tasks that none of them could tackle alone, such as end-to-end supply chain management.
  • Governance standards: bodies such as NIST (US) and the European AI Office are developing evaluation and certification frameworks for.

The future of AI agents

As intelligent agents gain prominence, businesses must adapt to a rapidly evolving landscape. Ethical concerns, transparency, and regulatory oversight will play critical roles in shaping the future of AI. As Amatriain pointed out, "This is a brave new world, and it’s happening now." The challenge for businesses and policymakers alike is to harness the power of intelligent agents while ensuring they serve humanity’s best interests.

The question is no longer whether artificial intelligence agents will transform business. The evidence from banking, law, logistics and healthcare shows that they already are. What executives, regulators and professionals need to ask now is how to design that future responsibly.

Frequently asked questions

What are AI agents?

AI agents are artificial intelligence systems designed to perceive their environment, reason about it and act autonomously to complete complex tasks. Unlike chatbots or conventional generative AI, they don't just respond; they plan, use external tools and chain actions together to reach a defined goal without constant human supervision.

How does an AI agent work?

An AI agent operates in a continuous cycle: it perceives information from its environment, processes it with language models or reasoning algorithms, makes a decision and carries out a real action (such as sending a message, querying a database or generating a report). The result of each action feeds into the next cycle, allowing the agent to adapt to changes in its context.

What is the difference between an AI agent and a chatbot?

A chatbot answers a single question and stops. An AI agent goes further: it can chain multiple actions, use external tools (web browsers, APIs, databases) and pursue a goal autonomously over time. The key difference is the ability to plan and act, not just generate text.

What are some intelligent agent examples in business?

The most notable intelligent agent examples are customer service agents that resolve complex issues autonomously, financial agents that monitor portfolios and detect fraud in real time, logistics agents that optimize routes on the fly, and legal agents that analyze contracts and assess regulatory risks. In every case, the agent acts, learns, and improves with each cycle.

How does the evaluation of AI agents work?

The evaluation of AI agents considers three main dimensions:

  1. Traceability: can the decisions made be explained?
  2. Alignment with the organization's values: does the agent act within the defined ethical and legal limits?
  3. Effectiveness: does it achieve its goals with the expected accuracy and cost?

In high-risk contexts, the EU AI Act requires a conformity assessment before deployment, carried out by an independent body in certain cases.

How are AI agents evolving in 2025-2026?

The evolution of AI agents in 2025-2026 is shaped by three trends: multimodal agents that process text, voice, and images in an integrated way; collaborative multi-agent systems that coordinate complex tasks with one another; and specific regulatory frameworks for their evaluation and certification in critical sectors.

All written content is licensed under a Creative Commons Attribution 4.0 International license.