Trends in Artificial Intelligence

Artificial intelligence (AI) is no longer an emerging technology but a transformative force redefining industries, organizations and societies at unprecedented speed. What seemed like science fiction just five years ago (systems capable of generating text, code, images, music or videos indistinguishable from what humans produce, officially passing the Turing test) is now an operational reality in millions of companies: according to the Stanford AI Index, 78 % of organizations were using AI in at least one business function by 2024, up from 55 % the previous year.

The speed of change doesn't stop: every month new capabilities emerge, every quarter boundaries that seemed far away move, unit costs of AI plummet, and every year it forces us to rethink what we thought we knew about the future of work, competition and business strategy. In the words of Sam Altman, CEO of OpenAI.

«The cost of using a given level of AI drops approximately 10 times every 12 months [...]. Moore's law changed the world with a 2x improvement every 18 months; this is incomparably stronger.»

And according to Dario Amodei, co-founder and CEO of Anthropic:

«By 2026 or 2027, we will have AI systems that will be, generally speaking, better than almost all humans at almost everything.»

AI raises strategic questions that go beyond technology and affect strategy, organization, and people: How can companies compete when innovation cycles are measured in months? How can organizations govern systems that evolve faster than their structures? How can they prepare people for jobs that do not yet exist? And how can they balance the speed of adoption with effective risk control?

But concrete operational dilemmas also arise: Is it better to invest in specific micro-tools that quickly solve bottlenecks, or to go for powerful multi-agent systems that promise consistency and organization-wide impact but require significant investment and carry the risk of rapid obsolescence? How can organizations conduct rigorous cost-benefit-risk analysis to prioritize among hundreds or even thousands of pilots? And how can they scale prototypes that work in controlled environments but, when deployed at real scale, encounter unexpected costs, emerging hallucinations, and support requirements that overload teams?

The experience of the past few years is beginning to reveal some patterns:

• Effective AI adoption is not just about acquiring technology or launching pilots: it requires organizational transformation, robust governance frameworks, ongoing training, and a deep understanding of technical, regulatory, and reputational risks.

• Organizations that move forward successfully are not necessarily those that invest the most, but those that best integrate technology, people, processes and control.

• The cost of inaction is no longer theoretical: the gap between pioneers and laggards widens exponentially, because every quarter of delay today can translate into years of competitive disadvantage tomorrow.

General productivity tools (securitized enterprise copilots) offer significant improvements immediately, provided they are accompanied by mandatory training and safe use frameworks, as required by European regulations. An OECD review of experimental studies shows substantial average productivity gains from the use of generative AI:

• In writing tasks, average execution time is reduced by 40% and quality increases by 18 %.

• In software development, programmers complete tasks 56% faster.

• In consulting, professionals using AI perform 12% more tasks, complete them 25 % faster and achieve more than a 40% improvement in quality.

• In customer service, professionals supported by AI assistants resolve 14 % more incidents.

The real bottleneck, therefore, is not technical but organizational. Technology is moving faster than internal structures. Friction emerges in slow processes, poorly governed data, overcrowded committees, diffuse responsibilities, and bureaucratic approval cycles. Organizations that do not redesign their internal machinery to enable speed without losing control will be unable to capture the value of AI, no matter how sophisticated the technology they employ. As Gartner puts it:

«The enormous potential business value of AI is not going to materialize spontaneously. Success will depend on closely aligned pilots [with the business], proactive infrastructure benchmarking, and coordination between AI and business teams to create tangible business value».

Two conditions underpin all of this:

• The value that an AI system delivers depends fundamentally on the quality of the data on which it is trained and the quality of the data it operates on in practice: a conversational assistant trained on outdated internal documentation will consistently reproduce those inaccuracies in every response.

• The quality of what a model produces depends largely on the quality of what it is asked to do. The ability to define the problem, provide relevant context and set precise constraints is not a niche technical skill; it is the new operational literacy, and the gap between those who master it and those who do not translates directly into a productivity divide.

Finally, it is critical to manage expectations: AI certainly brings real and measurable value, but it does not immediately replace critical processes or solve structural problems on its own.

This paper presents 22 key trends in AI, ranging from capabilities that are already operational to emerging developments that are driving strategic decisions today. It is not intended to be a technical manual or a speculative long-term projection, but rather a rigorous analysis of what is already happening and what is about to happen, designed for decision makers in complex and regulated environments.  

It is structured into four sections:

• The Technological Explosion of AI examines capabilities that are already operational and transforming organizations, from the democratization of multimodal generative AI to the rise of agentic systems, including Machine Learning accelerated by generative AI, new approaches to software creation such as vibe coding, and the integration of AI into robotics and physical systems. 
• AI Risks, Regulation, and Safety addresses the critical challenges of rapid adoption, including technical, legal, and reputational risks; the evolving regulatory frameworks and standards; the emerging conflict between defensive and adversarial AI in cybersecurity; inherent AI vulnerabilities; and ongoing tensions around privacy and intellectual property. 
•  AI Governance and Impact on People focuses on how organizations are responding structurally: creating AI-specific corporate governance models, industrializing deployment through advanced operational practices (MLOps, LLMOps), transforming professional roles and profiles, driving AI adoption across sectors (AI + X), addressing sustainability and social impact, and implementing operational ethical frameworks that go beyond statements of principle; and also addresses the impact of AI on people's daily lives. 
•  Frontiers of AI analyzes emerging developments already shaping strategic decisions: the geopolitics and technological sovereignty of AI, AI-first and AI-only organizations, AI-assisted scientific research, digital twins and simulations of human behavior, Ambient AI and invisible computing, interactions between AI and quantum computing, and artificial general intelligence (AGI) as a strategic horizon that can no longer be ignored.

Finally, a case study is presented: GenMS™ Sybil, a conversational assistant trained using this very paper and developed in a single day. It enables interactive exploration of the trends discussed and illustrates in practice the concepts, architectures, and controls highlighted throughout the document.

This paper is not intended to be exhaustive - AI is evolving too rapidly for that - but rather to provide a solid conceptual framework, concrete examples, verifiable references, and decision-making criteria for navigating an environment of accelerated change with rigor, realism, and responsibility.