Lilian Delaveau - Speaker on artificial intelligence

AI-assisted development tools accelerate code production and change the way engineering tasks are approached. But writing faster is not enough: the team must continue to understand its product, test its changes, and master its architecture.

This conference explains how to integrate AI into the practices of a development team. It addresses delegation, code review, testing, documentation, security, and technical debt. The tools are placed back into a delivery process that developers and their managers maintain control over.

We look at the decisions that remain necessary when AI proposes an implementation: clarifying the need, understanding the changes, verifying expected behaviors, and assessing their consequences on the rest of the system. The role of tech leads and engineering standards provide a framework for this acceleration.

The goal is to benefit from AI-assisted development without compromising maintainability or losing code understanding. It is about linking the production pace to the quality, security, and operational requirements of the product.

The presentation is aimed at CTOs, VP Engineering, tech leads, developers, IT service companies, and software publishers. Formats: 45 to 60-minute conference or two-hour event masterclass, with possible extension into team training. French or English. In-person in Brittany, Île-de-France, and Centre-Val de Loire; remote possible. Catalog: https://liliandelaveau.fr.

AI agents promise to respond to customers, manage workflows, and take on increasingly lengthy tasks. But what does this word really mean? An augmented assistant, automation, an autonomous system, or a new marketing label?

This conference provides benchmarks to understand the differences between assistant, automated workflow, agent, and autonomous system. It helps to look at what the system can actually do, what information it uses, and what actions it is allowed to perform.

We address data access, action rights, human supervision, and safeguards. The topic is related to the uses of the company and the constraints of its information system: a successful demonstration is not enough to establish that an agent can intervene in all processes.

The goal is to understand what an AI agent can really do, what needs to be framed before delegating a task to it, and how to spot promises that exceed the system's capabilities. The discussion connects technical possibilities and organizational decisions.

The presentation is aimed at CIOs, CTOs, business leaders, software publishers, and innovation teams. The level can be adjusted for a leadership perspective or an architecture and risk angle. Basic format: 60-minute conference, in French or English. In-person in Brittany, Île-de-France, and Centre-Val de Loire; remote participation possible. Catalog: https://liliandelaveau.fr.

How to establish useful rules on artificial intelligence without turning the subject into a taboo? Between improvised uses with confidential data and a caution that prevents any experimentation, the company needs common benchmarks and proportionate decisions.

This conference connects AI governance to the concrete uses of leaders, technical teams, and professions. It addresses sensitive data, SaaS tools, security, human validation, and the distribution of responsibilities. The AI Act and data protection issues are discussed in relation to technical choices and organizational practices.

The goal is to establish clear, actionable usage rules that are adapted to real risks. What can be entrusted to a tool? What data to use? When to foresee human validation? How to maintain control over access and actions, particularly when agents enter workflows?

The intervention helps to build a common discussion between business directions, IT departments, security officers, data protection officers, legal, and human resources. It invites distinguishing the capabilities of tools, their limits, and the conditions of their use, in order to move forward with clarity.

The content is tailored to the level and challenges of the participants. Basic format: 60-minute conference, in French or English. In-person in Brittany, Île-de-France, and Centre-Val de Loire; remote participation possible. Catalog and references: https://liliandelaveau.fr.

Why do generative AI projects get stuck after the demonstration? A prototype can convince a management committee while remaining difficult to exploit on a daily basis. Data, costs, security, evaluation, and team organization scale differently when the system must serve real users.

This conference provides a framework for transitioning from AI POCs to production. It connects technical choices to business decisions: which uses to prioritize, which data to make accessible, how to measure quality, where to place human validation, and how to track performance over time.

We address workflows, responsibilities, LLMOps, and team adoption. The goal is to know what to industrialize, what to stop, and what to frame before investing. An assistant, a RAG, or an agent is not judged solely on a successful demonstration: one must also understand its limits and the conditions of its exploitation.

The intervention is aimed at COMEX, CIOs, innovation departments, data teams, and business departments. The angle is tailored to the level of participants to make the trade-offs understandable and concrete.

Formats: keynote of 30 to 45 minutes, 60-minute conference followed by questions, or two-hour masterclass. Intervention in French or English. In-person in Brittany, Île-de-France, and Centre-Val de Loire; remote possible. Catalog: https://liliandelaveau.fr.