2025 marked a decisive turning point in the adoption of artificial intelligence in business. According to Lecko’s annual study on organizational transformation, we are witnessing a shift from experimentation to the structural integration of AI into daily workflows. Yet this transformation comes with major challenges and paradoxes that compel organizations to rethink their modes of operation at a fundamental level.
From the isolated tool to a coherent AI architecture
Organizations can no longer merely “choose an AI.” They must now design a coherent AI architecture capable of weaving together specialized models such as Claude, GPT, or Mistral, orchestration layers, and data connectors based on Retrieval-Augmented Generation (RAG) to guarantee the reliability of responses.
This evolution signals a new level of maturity: AI is no longer deployed piecemeal, but is part of a broader, overarching strategy.
The Rise of Agentic AI
Lecko highlights the shift from simple generative AI to agentic AI, a major evolution that radically transforms business processes. Unlike traditional assistance tools, an AI agent is capable of executing tasks autonomously, such as scheduling a meeting or handling an HR process, by interacting directly with the working environment.
This autonomy rests on three essential capabilities: the ability to analyze complex situations, the development of strategies, and direct interaction with existing tools like CRMs, ERPs, or calendars. Agentic AI aims to fully substitute tasks, and no longer merely improve or facilitate them.
Job-by-job specialization
AI is now integrated at the heart of department-specific processes.
In human resources, solutions like Workday Illuminate deploy specialized agents to automate recruitment or draft contracts. ServiceNow Now Assist automates incident summaries and the generation of workflows for IT teams. Sales teams benefit from agents capable of qualifying leads or autonomously preparing proposals.
In 2026, value shifts toward agentic platforms capable of orchestrating multiple agents simultaneously. Vendors such as Jalios, Jamespot, LumApps, or Microsoft offer low-code studios enabling business users to build their own bespoke agents. “Meta-agents” or AI concierges direct queries to the most relevant agent, querying multiple databases at once.
According to Gartner, by 2028, 15% of daily business decisions will be made autonomously by AI agents. This outlook underscores the scale of the ongoing transformation.
The Productivity Paradox
One of the study’s most important findings concerns what Lecko calls the “productivity paradox.” Contrary to expectations, AI does not automatically improve work rhythms. To achieve a real gain, the focus must be on substitution tasks, where AI actually replaces a human action, rather than on comfort or augmentation tasks.
If the time savings promised by AI are not used to substitute recurring tasks, they remain invisible at the organizational level and can even fuel hyperconnectivity. AI acts as a revealer: it can only transform processes if the documentary heritage is sound and well-structured. It becomes ineffective if business rules are unclear or data are poorly organized.
Fighting organizational noise
Organizational noise, defined as the surplus of unfiltered and unprioritized digital solicitations that fragment attention, is one of the scourges of the modern workplace. Agentic AI offers several levers to combat it.
Systems like Jamespot’s Teamwork Assistant provide intelligent summaries of notifications and prioritization of messages. Tools such as Staffbase or Sociabble use AI to segment audiences and tailor feeds, avoiding the spread of irrelevant information to field employees.
The emergence of orchestration meta-agents enables routing requests to the right specialized agent, avoiding unnecessary back-and-forth within the information system. Platforms such as LumApps with its Agent Hub or Elium centralize knowledge to respond in natural language, reducing the noise from unsuccessful searches.
The agentic ecosystem is also becoming more attractive by offering innovations such as “zero-file” or knowledge structuring, which structurally reduces the central role of email, the primary vector of information pollution in the enterprise.
Meeting optimization: promises and limits
AI can contribute to curbing meeting inefficiency, albeit with important caveats. Current tools like Copilot, Leexi, or Fireflies excel at transcription, generating minutes, and extracting action plans. By providing pre-summaries, AI helps shorten meetings and reduce the number of participants required.
However, the study highlights a major technical limitation: hybrid meetings are particularly difficult for AI to handle, as it struggles to correctly identify in-person attendees versus those online.
Moreover, AI acts as a mirror of organizational effectiveness. A poorly structured meeting or one without a facilitator will yield a mediocre synthesis. Deploying the tool alone is not enough to improve work rhythms; it must be accompanied by questions about the very necessity of the meeting.
The risks and limits of generative AI
The study raises several significant risks.
Paradoxically, AI can worsen information pollution if it is not accompanied by changes in practices. The ease of production can further saturate the digital environment, creating a conveyor-belt effect where an increasing share of content is generated automatically without necessarily adding value.
The phenomenon of “Shadow AI” is also worrisome: about 49% of corporate users rely on AI without informing their supervisors, which contributes to fragmenting the Digital Workplace and complicates the control of information flows.
Studies also warn of declines in users’ cognitive and memory abilities. The presence of AI can lead participants to pay less attention, relying on automatic summaries to catch up on what they missed, an effect described as “sleep at the wheel.”
Finally, the environmental impact is colossal: electricity consumption of data centers could double by 2026, raising urgent questions about the sustainability of this technological revolution.
