Secure gpt: Veolia’s transition from generative AI to agentic AI
In a world where digital transformation has become inevitable, Veolia stands out with its bold approach to generative and agentic AI. Since its launch in September 2023, the Secure GPT application has perfectly illustrated this transition, symbolizing not only technological innovation but also a strong commitment to a sustainable future. In this article, we will explore the key milestones of this transformation, the challenges encountered, and how the integration of smart solutions has become essential for the circular economy.
The Birth of Secure GPT: From Generative AI to Agentic AI
Veolia’s journey toward agentic AI truly took off with the launch of Secure GPT, designed to bring together an ecosystem of intelligent agents and assistants to serve employees. As of January 2025, this system already has more than 214 AI assistants, used daily by 58,000 employees across 41 entities. This impressive figure demonstrates the group’s commitment to leveraging new technologies to improve its internal operations. A Vision Focused on Digital Transformation
The transition to an AI-powered infrastructure at Veolia is part of a broader digital transformation effort initiated in 2013. The company, which now has more than 218,000 employees, led the way with the adoption of Google Workspace, followed by key integrations with Google Cloud and AWS. Over the years, Veolia has implemented a cloud-first strategy, prioritizing online services to improve its operational efficiency.
This strategy has fostered the emergence of generative AI, embodied by ChatGPT, which has enabled the company to seize opportunities for innovation while being aware of challenges, such as security risks and bias. This dynamic has reinforced the idea that AI is a powerful tool, capable of optimizing processes while respecting sustainability values.
A solid infrastructure: the Secure GPT architecture
From the outset, the Secure GPT architecture was designed “
by design » to integrate different language models. Whether answering specific questions or translating documents, each assistant is configurable according to users’ needs. Thus, more than twelve LLMs (Large Language Model) are available, each one selected for its ability to handle varied use cases. LLM for Conversation
- LLM for Translation
- LLM for Querying PDF Files
- During the widespread deployment phase in October 2023, a well-structured API was put in place to ensure access to IT teams across 57 countries. This ability to quickly deploy tools streamlined the process and propelled the organization’s efficiency to new heights.
A focus on sustainability and innovation
The transition to agentic AI is not just about optimizing internal processes, it is also part of a broader sustainability vision. Veolia is actively engaged in its approach to
sustainability by highlighting the use of less energy-intensive AI models as part of their AI for Green Up program. This initiative aims to minimize the environmental costs inherent in the use of artificial intelligence while maximizing benefits. Fouad Maach, head of IT architecture and engineering, emphasizes that the objective is to
increaseemployee capabilities rather than replacing them, illustrating a clear commitment to respecting the environment while fostering innovation. The Evolution of Assistance Systems at Veolia
In January 2025, Veolia recorded high usage of its assistance solutions, demonstrating the exponential growth curve of smart technology integration. But how has this evolution translated concretely into the daily lives of teams?
AI Assistants: A Response to Diverse Needs
The various AI assistants developed within Secure GPT can be classified into three levels of complexity. This classification provides a better understanding of their capabilities and scope:
Level
| Description | Examples of Use | Level 1 |
|---|---|---|
| LLM with a customized prompt to address a specific case | Basic user assistance | Level 2 |
| Use of RAG (Retrieval-Augmented Generation) | Querying complex documents | Level 3 |
| Ability to process structured and unstructured data | Interactions with the corporate information system | A Level 3 assistant might, for example, interact with the corporate information system to submit purchase orders or send emails. These tools are essential to Veolia’s infrastructure, facilitating information exchange and strategic decision-making. |
The Importance of Culture and Training
Another fundamental aspect of Secure GPT deployment is employee training. Between September and November 2023, an acculturation program was implemented to educate employees on the capabilities of generative and agentic AI. A team of 400 AI champions was formed to train more than 13,000 employees on the use of these new tools, which facilitated the acceptance and integration of these smart solutions.
The training not only involves teaching the basics of AI, but also reviewing operational processes. This allows the data and digital teams to co-construct solutions aligned with the group’s circular economy objectives, whether reducing waste or improving energy performance. A growing trend toward agentic AIThe rise of agentic AI at Veolia offers a fascinating glimpse into the future of intelligence within large organizations. Starting in June 2024, experiments have been launched to create an agentic platform that connects multiple assistants and agents, each with specific roles. These agents can interact, exchange information, and integrate seamlessly with existing IT systems. This multi-agent workflow model is expected to revolutionize working methods within the company.
Challenges to overcome and successes to celebrateAny transformation, even a promising one, presents its share of challenges. For Veolia, managing the risks associated with generative AI remains a priority, particularly with regard to data security and protection against bias. As technologies advance, the company continues to evaluate best practices to minimize these risks while leveraging AI’s capabilities to the fullest. Identifying and Minimizing Risks
Fouad Maach raised major concerns such as data exfiltration, intellectual property infringement, and AI model hallucinations. Analyzing these issues is fundamental to ensuring not only the company’s protection but also user confidence in these technologies. Veolia is therefore working to develop robust mitigation strategies, ensuring that generative and agentic AI are compatible with its values of sustainability and ethics.
Building Performance Through Innovation By embarking on this path, Veolia is not only adopting technologies: it is creating a culture of innovation tinged with responsibility. By integrating generative and agentic AI, the company ensures it remains at the forefront, not only for its teams, but also for its customers and the environment. This entire process would lay the foundation for an intelligent support network, guaranteeing effective solutions to complex challenges related to natural resources and waste management. A vision of the future driven by innovation and collaboration Veolia embodies a vision of the future where technological innovation and sustainability are not opposing factors, but complementary. Through its Secure GPT platform and its active approach to energy and the environment, the company is working towards a world where generative and agentic AI transforms resource exploitation into a circular economy model.
Finally, as technological solutions continue to evolve, the key to success lies in a proactive approach, combining training, collaboration, and constant evaluation of results. By maintaining this balance between advancing technologically and remaining anchored in the core values of sustainability and ethics, Veolia is ideally positioned to face future environmental and sustainable development challenges.
Catégories : Non classé
Tags :