Based on the report “Big Data & AI Insiders — AI Agents: Will the Revolution Happen?” published in 2025 by Big Data & AI Paris trade show (RX France).
A Technological Promise Seeking Maturity
In 2025, the word “agent” no longer refers to an employee, but to software. The AI agent embodies a new frontier: an autonomous program capable of reasoning, triggering actions without human intervention, and interacting with other agents in a distributed system. The idea fascinates, sparks dizzying projections—but raises questions.
Gartner predicts that 15% of business decisions will be made by agents by 2028. Goldman Sachs mentions 300 million jobs potentially affected. And PwC announces a gain of $15.7 trillion globally. Yet these figures still remain scenarios. The Big Data & AI Insiders report (2025) calls for nuance: “Will the revolution happen? It’s up to each person to form an opinion, in light of the perspectives we invite you to explore here.”
Assistants, Copilots, Agents: Clarifying Roles
The AI agent should not be confused with a simple conversational assistant. Unlike copilots (such as Copilot, Gemini, or Claude), the agent takes initiative, acts autonomously, and can interact with information systems to execute tasks. The report mentions concrete examples: an agent that identifies a product failure reported by a customer and directly orders the part in the IT system.
This paradigm shift relies on complex architecture: LLM, memory system, reasoning engines, and the ability to execute actions in third-party systems. We’re moving from generative AI to agentic AI, that is, from content generation to decision and action generation.
Between Skepticism and Long-Term Strategy
But the terrain remains uncertain. Stéphane Bout (McKinsey) reminds us: “90% of AI projects in 2024 didn’t go beyond the experimentation stage.” Companies, burned by PoCs without clear ROI, are proceeding cautiously. Chadi Hantouche (Wavestone) adds: “It’s a global transformation of work methods that’s beginning.”
Some industries (finance, energy, aerospace) are more reluctant, due to regulatory risks or predictability needs. Others, more tolerant of errors (retail, marketing, customer support), are tentatively engaging. The primary challenge remains governance: who pilots, who supervises, who validates decisions made by an autonomous system?
Multi-Agent Systems, the New Organizational Frontier
The future isn’t made of a single agent, but of an interconnected network. In a multi-agent system, each specialized unit dialogues with its neighbors via LLMs, transmits instructions, and triggers chains of microtasks until a result is obtained. This algorithmic orchestration prefigures more fluid organizations—but also more opaque ones.
Supervision becomes central. Some speak of “super-agents” or “orchestrator agents,” capable of monitoring, arbitrating, and interrupting action chains. The architecture is inspired by the “human-on-the-loop” logic: maintaining humans as actors of last resort. A cautious design, inherited from lessons learned from generative AI.
Interoperability and Sovereignty: Conditions for Scalability
The market is organizing around open standards. In November 2024, Anthropic proposed the MCP protocol to connect agents and enterprise systems. In April 2025, Google launched A2A to enable agents to communicate with each other. These initiatives received support from major players (OpenAI, Microsoft, SAP, Salesforce). The challenge: avoid market fragmentation and allow agents to adapt to existing systems.
But this interoperability raises other questions: who governs these standards? Who guarantees their security, scalability, compliance with European regulations? The sovereign temptation is gaining ground. Players like LightOn, Ionos, or Mistral are defending AI hosted in Europe, on private cloud, with strong guarantees regarding confidentiality.
The Question of Cost and Business Model
The AI agent isn’t free. It requires powerful infrastructures, trained models, supervision systems. Scaling encounters high costs. Hence reflection on billing: by usage? by performance? by business impact? No consensus emerges.
The risk of a rebound effect is real. As Big Data & AI Insiders explains, “the autonomous nature of agents adds an opaque dimension to resource consumption.” Monitoring and optimization mechanisms are essential to avoid excesses, especially in multi-agent contexts.
The Test of Reality: Use Cases and Instructive Failures
Some use cases are emerging: HR onboarding, marketing campaigns, fraud detection, regulatory summary generation. But large-scale production deployments are rare. The Klarna example is revealing: by replacing part of its customer service with agents, the company saw response times improve… but customer satisfaction plummet. Two years later, partial return to human agents.
This failure highlights a key point: generative AI isn’t always ready to be substitutive. Agentic AI might be… but its adoption will depend on its actual relevance in business processes.
An Agent-Driven Company Implies an HR Revolution
The arrival of agents disrupts organizational structure. What place for an agent that’s a twin of an employee? Should we create IT/business pairs, as in the Big Data era? How to evaluate an agent’s performance? Who is responsible in case of error? So many unanswered questions, but they prefigure new roles, new evaluation frameworks, and probably a redesign of certain support functions.
Reuters estimated in 2024 that 70% of workforces needed “upskilling” to remain competitive against AI technologies. Training becomes a strategic challenge, as much as adoption. And the market, in return, values no code/low code interfaces, to allow business teams to design their own agents with little technical expertise.
Ethical and Societal Challenges Lurking
Behind the technicality lies a broader debate: what work world do we want to build? The AI agent, by delegating cognitive load, lightens the human burden. But at what cost? Loss of meaning? Obsolescence of know-how? Questioning of professional autonomy? The report mentions a worrying paradox: what if, tomorrow, agents themselves became obsolete?
DBS Bank in Singapore eliminated 4,000 positions in 2025 by integrating agents. A controversial decision, even though the company announces the creation of 1,000 new jobs. Europe, more cautious, is watching. But signals are accumulating. The promised disruption is profound.
A Revolution, Yes, but Conditional on Value
The closing article of the Insiders report concludes bluntly: “Yes, the revolution will happen.” But it immediately nuances the scope: speed, intensity, and acceptance will depend on companies’ ability to create value. The AI agent doesn’t impose itself through its technology, but through perceived utility. Acculturation, governance, trust, sovereignty: these are the real levers.
The first weak signals will be decisive. They will tell whether agents keep their promises or lock themselves into the same dead ends as generative AI: PoCs without follow-up, tools without adoption, hype without strategy. Lucidity then becomes the best compass. Strategic, technological, human lucidity.
