The arrival of GPT-5 marked a change in perspective within companies. Not because the model is disappointing, but because it reveals a structural limitation: the rise of models no longer automatically translates into greater value.
Advances in reasoning, multimodality, and automation are opening up new perspectives and marking a turning point. The transformation of uses, organizations, and operational performance is now being built into the way economic players integrate models into their processes and systems. The plateau observed is not technological; it is strategic.
When power is no longer enough
GPT-5 highlights a profound shift in the issues at stake. The debate is no longer about the intrinsic performance of models, but about the ability of organizations to leverage them to create concrete and sustainable value. In many projects, the same pattern emerges: convincing demonstrations, followed by complex production rollouts, mixed results across different business lines, and increased dependence on generalist models whose behavior remains difficult to control.
Despite their level of maturity, models remain difficult to integrate on a large scale within organizations. Nearly four out of five AI projects, or about 80%1, fail to meet their objectives or are abandoned along the way. This failure rate is almost twice that of traditional IT projects.2 Generative AI, long perceived as a natural accelerator of transformation, is becoming a demanding tool to manage. Senior management is therefore shifting its attention to more structural issues: operating costs, operational risks, robustness of responses, and the ability to demonstrate a tangible return on investment.
For companies, value creation is no longer about choosing the “best” model or optimizing prompts. It depends above all on the structuring of uses. This evolution is redefining expectations of Large Language Model (LLM) providers, who are now expected to be able to serve concrete and operational cases. The example of Claude, from Anthropic, which specializes in IT development, illustrates this transition towards models designed to integrate effectively into specific business contexts.
Moving away from the illusion of the central model
This plateau marks a stage of maturity. It invites companies to move beyond their fascination with raw performance and refocus on the essentials: concrete applications, robust systems, and a lasting relationship of trust.
In many organizations, however, AI strategy remains centered around a central model that is presented as capable of understanding, deciding, and producing everything. Today, this vision is showing its limitations and hindering long-term value creation. Value is now built into the architecture that frames AI. Knowledge structuring, clear delineation of uses, specialization by business case, and orchestration of rules, data, and generation are becoming key performance factors. Generative AI is thus a building block within a larger system designed to ensure consistency, control, and reliability.
The most successful companies will be those that know how to use AI systems that are useful before they are spectacular, controlled before they are autonomous, and sustainable before they are simply innovative.
2026: AI enters the era of results
By 2026, AI will have fully entered a phase of economic and operational maturity. The gradual implementation of the AI Act 3 will accompany and accelerate this movement. By imposing increased requirements in terms of understanding, control, and explainability, it will push companies to structure systems that are more transparent, better governed, and closely aligned with business challenges.
In this context, organizations are gradually moving away from experimentation and focusing their investments on uses that can produce measurable and industrializable results. AI is becoming integrated into core processes: it automates task chains, supports decision-making, and works closely with human teams. It is becoming a key driver of operational transformation.
To meet these new performance and control requirements, technological strategies are diversifying. Organizations are combining large models with more specialized models to optimize costs, energy impact, and business alignment. This approach requires careful orchestration, in which human expertise plays a central role in framing uses, supervising decisions, and ensuring the reliability of the systems deployed.
The future of AI will depend on the ability of organizations to transform these technologies into sustainable levers of performance and trust. This value will be based on a balance between companies that are able to structure specific business uses and model publishers that are able to evolve their models towards truly operational specialization.
1 Source: RAND think tank report
2 Source: Gartner study
3 AI Act: European regulation on artificial intelligence that sets out obligations for suppliers of general-purpose AI systems (GPAI) and prohibits certain AI systems that infringe on fundamental rights.
Opinion piece by Samir Dilmi published on économie matin.
Share this article!
