Interview on AI Agents with Hanan Ouazan, Managing Partner and Global Lead AI Acceleration at Artefact

Hanan joined Artefact at its very beginning 10 years ago and has always been passionate about AI technologies. He plays a key role in the development of generative AI solutions at Artefact for our customers, harnessing constantly evolving advanced technologies to better meet their needs according to their industry imperatives.

A graduate of Centrale Lille and the École Normale Supérieure (ENS), he completed his MVA (Mathematics, Vision, Learning) research master’s degree at the ENS, one of the most advanced programs in Machine Learning, Data Mining, and Statistical Learning. Hanan leads a team of talented engineers in the design, development and industrialization of artificial intelligence use cases.

AI agents are no longer simply tools performing automated tasks. Unlike RPA (Robotic Process Automation), which executes predefined tasks in a precise sequence, AI agents can adapt to their environment, making decisions and optimizing processes without human intervention.

This marks a real step towards more autonomous, flexible and intelligent AI. These AI agents bring a new, more adaptive approach because they can:

• Observe their environment and adjust their behavior according to the data.
• Interact with multiple systems to perform complex tasks.
• Optimize and automate processes without constant human intervention.

“AI agents and Agentic will be the most talked-about technological terms in 2025. This evolution in automation takes advantage of the exceptional performance that generative AI has brought. I like to call it RPA 2.0, capable of integrating into dynamic workflows and going beyond the simple role assistant.”

Hanan Ouazan, Managing Partner Artefact & Global Lead AI Acceleration

RPA (Robotic Process Automation) technology consists in automating a process by breaking it down into a succession of tasks, which are chained together until the result is obtained, often using a simple process: Read, Extract, Write.

Generative AI technologies consist of three tool elements:

1. LLM: First, we had models that allowed us to answer questions based on their knowledge.
2. Assistant: Then came the time of assistants capable of interfacing with knowledge systems to help us in our daily lives (RAG use, etc.) – the famous “COPILOT”. Here, AI doesn’t “execute” anything but rather enriches its knowledge and assists its user.
3. Agent: An agent is an application that seeks to achieve a goal by observing the world and acting on it with the tools at its disposal. What’s new is this “awareness” of the environment, which enables the agent to adapt to unforeseen events.

And that’s where it contrasts with the notion of RPA. It’s the agent’s ability to adapt to the unexpected that often earns it the label RPA 2.0!

AI agents have applications in many fields.

• Customer relations: intelligent chatbots capable not only of answering questions, but also of performing actions (updating a contract, generating an invoice, handling a specific request).
• Back office: EDM 2.0, automated document management, incident handling and internal requests.
• Code writing: AIs assist developers by generating code, optimizing scripts or automating tests.

 

“Use cases are going to have to be rethought in terms of process and not just translated into AI. The aim is not to superimpose AI, but to redefine the process without the constraints that may have existed.”

Hanan Ouazan, Managing Partner Artefact & Global Lead AI Acceleration

 

“We’re seeing the emergence of AI tools capable of generating code, deploying it, and even automating complex processes. These technologies are profoundly disrupting software development, making it much faster and more accessible to evolve applications.”

Hanan Ouazan, Managing Partner Artefact & Global Lead AI Acceleration

AI Agents have two major uses:

1. Rapid bootstrapping (e.g. Bolt): immediate transformation of concepts into prototypes.
2. Development assistance (e.g. Copilot, Cursor): code completion, refactoring, test generation.

However, this evolution does not replace human expertise. These tools require rigorous supervision to avoid structural errors, and in particular the risk of “house of cards” code, which may appear high-performance but is unstable in production.

Today, the connection between an AI agent and software is mainly based on service-specific APIs. To communicate with systems such as Google Drive, Slack, and Confluence, for example, this involves:

1. Integrating three distinct APIs, each with its own technical specificities.
2. Configuring the AI agent with three connectors, each requiring individual authentication and configuration.
3. Launching three separate queries, whose answers are then aggregated or synthesized by the agent.

This process is not only complex, but also difficult to maintain and evolve, especially on a large scale. This is precisely the problem that the initiative launched by Anthropic, with the Model Context Protocol (MCP) aims to solve. This new protocol acts as a universal standardized API, specifically designed to simplify interactions between AI agents and software.

Finally, it’s important to understand that AI Agents call into question the structure of SaaS software, which is based on three fundamental building blocks (UX, Intelligence,Data). This technology calls these 3 components into question:

1. UX: With AI agents, user interaction is evolving. Prompts and automation replace the traditional click, making classic UX less central. The experience is redefined around invisible workflows or workflows adapted to the needs of the agents.
2. Intelligence: Automations often involve actions between several systems. Open, standardized logic could replace workflows locked away in isolated ecosystems.
3. Data: Data ownership remains an obstacle to full integration.

 

“Today, SaaS publishers integrate layers of automation into their platforms, but this approach is showing its limits in the face of multi-platform needs. The future of SaaS probably lies in open, interconnected and adaptive logics, at the service of businesses.”

Hanan Ouazan, Managing Partner Artefact & Global Lead AI Acceleration

The first AI agents, such as Copilot from Microsoft, Google and Salesforce, are becoming more widespread, but remain limited to simple tasks. More advanced agents are struggling to be industrialized, due to a lack of control and trust, while autonomous agents remain experimental.

A few months ago, at the Consumer Electronics Show (CES) in Las Vegas, Nvidia CEO Jensen Huang declared that “IT will become the HR of AI agents”.

According to Hanan Ouazan, while this vision is interesting and challenging, it is neither entirely realistic nor desirable.

In his view, IT will play a key role in the deployment of these AI agents but will not be able to take on the management of these tools alone. It will require a broader organizational transformation and the indispensable involvement of the business lines:

• A business challenge above all: AI agents automate business processes. Supervising them will require the involvement of operational and data teams.
• A profound organizational change: IT will be responsible for technical maintenance, but governance and agent optimization will be the responsibility of the business lines.

Generative AI is already transforming businesses: less execution, more supervision and optimization. As with industrial automation, skills must evolve to integrate these tools effectively.

 

“Automation alone isn’t enough: we already need to rethink business processes to avoid unnecessary complexity. AI should not be just another layer, but a lever for imagining more efficient models adapted to business needs.”

Hanan Ouazan, Managing Partner Artefact & Global Lead AI Acceleration