Conferences

For years, we pretended this was normal: the business talks to the product manager, who translates to the designer, who churns out static mockups, which the developer translates back into code. An endless cascade of translations, and a mountain of hidden costs everyone has learned to ignore.

AI is reshuffling the deck, but not where you’d expect. The real point isn’t to spit out throwaway prototypes that end up in the trash. It’s to design in an isolated environment that’s genuinely wired into the production codebase. No more mockups that lie: time for iteration that speaks the same language as the final code.

And from there, everything shifts. The bottleneck has moved: producing has become fast, it’s deciding that slows things down. Business teams and PMs test their ideas live, designers stop pushing pixels and reclaim their role as captains of the user experience, and developers finally get something concrete instead of specs to interpret. Roles are being redrawn. So is the organization.

What’s left are the real questions, the ones we rarely ask in time: what do we keep under our control, what do we hand off to the machine, and on what, or whom, are we willing to become dependent to get there?

“Open-source is dead”, according to X/Twitter. After all, why would we need to install tons of dependencies (with vulnerabilities and supply chain attacks ramping up), when our AI agents can re-implement everything for us?

To be fair, rumours of its death have been strongly exaggerated: the world still needs open-source more than ever. But the role of OSS maintainers is changing. From floods of AI slop PRs to newfound superpowers with fine-tuned agentic workflows to help with triage and maintenance, what does the job of maintainers look like at the age of AI?

Retour d’expérience sur l’introduction de l’IA dans une direction de la Région Normandie : de l’enthousiasme initial aux limites des POC, en passant par la Shadow AI, jusqu’à la mise en place d’une méthodologie agile orientée sur les besoins du métier, qui permet l’appropriation rapide par les agents.

Déléguer un projet critique à un expert est censé garantir la tranquillité d’esprit, c’est donc tout naturellement que j’ai décidé d’engager une “Wedding Planner as a Service” pour la plus grosse mise en prod de mon année, et peut-être de ma vie : mon mariage. En revanche, ce que je ne savais pas, c’est que ce projet s’est révélée être une boîte noire et un véritable SPOF. En voyant les mois et les semaines passer sans grand avancement et surtout, sans la gestion de projet que l’on m’avait vendue, j’ai dû procéder à une ré-internalisation de l’architecture de mon projet, en plein vol.

À travers ce retour d’expérience, je vous propose un parallèle entre mon projet personnel et nos métiers de la tech, entre SPOF, observabilité et gestion de crise.

En deux ans, l’IA s’est invitée à toutes les étapes de la chaîne de livraison logicielle : génération de code, revue de PR, tests, documentation, déploiement, monitoring. Les démos sont spectaculaires, les promesses vertigineuses, mais que se passe-t-il vraiment dans nos équipes une fois la hype retombée ?

Autour de la table, Chris, Lead Developer en ESN, François, mainteneur Open Source et Freelance, et Renan, Lead Tech en charge d’un produit.

Six mois après son passage à Codeurs en Seine, Quentin remonte sur scène pour faire le point sur la souveraineté numérique. L’IA rebat les cartes : où en est-on vraiment, et qu’est-ce qui bouge depuis ?

I’m a DevRel and a hobby pilot. One day I had a choice: attend a virtual tech conference or go flying. I chose both.

I built an AI version of myself, trained on my writing, talks, and technical knowledge and sent it to the conference while I was literally at 10,000 feet. It networked, answered questions, gave technical advice, and made commitments on my behalf.

When I landed, I had to deal with the consequences.

First Half (Accessible):

• Why I built an AI clone of myself
• The technical setup: voice, knowledge base, decision-making
• What it did at the conference (networking, technical discussions)
• The commitments my AI made that I had to honor
• The identity crisis: “Did I say that, or did my AI?”

Second Half (Technical Deep Dive):

• Architecture: Building a personal AI clone (LLMs, RAG, voice synthesis)
• MCP setup for conference integration (chat, networking, Q&A)
• What went right: Technical answers, networking, style matching
• What went wrong: Overcommitment, context misses, uncanny valley moments
• Live demo: My AI clone answering questions in real-time
• The philosophical problem: When is it me vs when is it just software?

Key Takeaways:

• How to build a personal AI assistant/clone (real architecture)
• The limits of AI representation (technical and social)
• When automation crosses into weird territory
• Practical lessons: What you should/shouldn’t delegate to AI
• The future question: If your AI can represent you, what’s left?

Vous avez lu “Design Patterns” du GoF ? Pas moi ! En revanche j’ai lu “A Pattern Language”, 1200 pages d’architecture qui ont inspiré le Gang of Four, avec l’intention de l’utiliser pour construire des maisons dans les Sims… Dans cette présentation, je vous propose de vous raconter ce que j’en ai tiré en tant que dev et en tant que linguiste. Une book review qui fait réfléchir à pourquoi on code :)

J’ai eu la chance de travailler avec des technologies avant gardistes, scala à partir de 2010, puis haskell, et désormais rust. Évidemment d’un point de vue de dev c’était cool, mais c’est toujours difficile à justifier au top management qui ne comprend pas la beauté d’un système de types expressif et nous embête avec des histoires de recrutement, de formation, de gestion d’équipe.

Un des textes qui revient souvent dans le domaine, c’est « beating the averages », un essai pour le moins auto-satisfait et suffisant, qui justifie les technos alternatives par une vélocité décuplée. Avec le recul de ces quatre expériences, je pense que cet essai n’est plus totalement d’actualité et j’aimerais le mitiger quelque peu.

Au programme :

• les impacts concrets sur ce qui fait peur au management :
• le recrutement
• la création et la gestion d’une équipe
• les impacts concrets pour les devs :
• l’impact d’un plus petit écosystème
• les gains à attendre
• les recettes qui ont plutôt marché et les écueils à éviter, en me basant sur des expériences moins heureuses dont j’ai pu avoir vent

Évidemment, on pourra conclure que « ça dépend », et surtout essayer de prendre du recul sur les critères de succès d’une telle entreprise.

I have to yell at my AI to get it to use the right context every time.

I added 15 “MANDATORY” labels in all caps to my skills. I ran the same prompt 10 times just to see if my skill would trigger consistently. And every time it failed, I pushed a commit to “fix it.”

I didn’t want a useful AI. I wanted an infallible AI.

I’m that person in a relationship who demands absolute perfection.

A mistake? Unforgivable. I want determinism.

I’m in a toxic relationship with AI.

And this toxicity has cost me time, energy… and way too many commits.

In this talk, I’ll share the story of this relationship in three acts:

• The honeymoon (10 min) — The early adopter’s excitement and the promises of context engineering
• The control freak (15 min) — The obsession with perfect context, the 15 “MANDATORY” labels in all caps
• Couples therapy (15 min) — The breakthrough: you can optimize your context in healthier ways. And AI isn’t a deterministic remote control—it’s an iteration partner that already makes me 3x more productive

I’ll show you excerpts from toxic skills, the successive diffs from my obsessive PR, and the healthy, efficient, maintainable version.

By the end of this talk, you’ll walk away with:

• Examples of context files that work — and ones that sabotage
• Warning signs of over-engineering — when your “MANDATORY” labels become counterproductive
• Best practices beyond context files — choosing the right interaction mode, managing the context window
• A realistic mindset — even with occasional misses, a well-calibrated context is still a valuable asset for building clean, shippable code faster

After building my own 8-bit CPU on a breadboard, along with a working VGA card and then a sound card, I’m here to present my 8-bit computer powered by the famous 65C02 — still on a breadboard. I’ll let you discover what it’s capable of on stage in a new session full of surprises.

We’ll see how, step by step, you can start from a simple transistor and end up with a computer worthy of Steve Wozniak’s Apple. We’ll also rediscover the early standards and protocols, like the well-known RS-232 (serial communication).

Retour aux bases d’OpenTelemetry : traces, métriques et logs, comment instrumenter proprement une application et démarrer sans s’y perdre.

Usuellement, des micro-services sont des processus isolés orchestrés sur un système d’exploitation complet. On nécéssite donc d’avoir une pile de composants et d’ingénérie pour décrire une petite API REST. Cette pratique, commune, bien que “relativement facile à mettre en place” augmente drastiquement la surface d’attaque potentielle et le temps de lancement de services, faisant du langage GO, un excellent candidat au développement de Microservices. Et si on poussait le vice en décidant de construire un système d’exploitation par services, en utilisant les capacité de virtualisation de notre processeur (son hyperviseur) pour construire des services isolés, tout petits et qui boot sacrément vite ?

Dans cet atelier, nous allons apprendre à créer des images spécialisées, des Unikernels, pour construire, à la demande, un système d’exploitation dont le rôle n’est que de faire tourner un seul service. Au moyen du langage OCaml et de la toolchain MirageOS nous construirons plusieurs petits services, simplement.

Au programme, virtualisation, appels systèmes, programmation fonctionnelle et système d’exploitation !

In this keynote, Marwen explores the concrete impact of AI on the UX design journey, from user research all the way to handoff. Between dramatically faster processes, new possibilities for exploration, and the automation of repetitive tasks, he raises the real question: are we witnessing a true revolution in the profession, or a powerful yet manageable evolution? A talk that takes a step back from the hype to examine AI’s promises, its limits, its risks, and the central role designers must continue to play in creating truly human experiences.

PHP a longtemps été pensé comme un langage strictement stateless : une requête, un processus, puis tout recommence. FrankenPHP change profondément ce contrat en permettant de lancer des workers PHP long-running directement au sein d’une application Symfony. Pas pour transformer Symfony en serveur Node, mais pour lui donner des capacités qu’il n’a jamais pu avoir jusque-là.

Dans cette conférence, je propose un nouveau pattern : les sidekicks applicatifs. Des workers PHP spécialisés, hors cycle HTTP, qui écoutent leur environnement et reconfigurent l’application en temps réel — sans polling, sans TTL approximatifs, sans redeploy.

À travers des cas concrets (découverte Redis Sentinel, rotation de secrets et de clés JWT, feature flags dynamiques, invalidation ciblée de cache ou réactions de sécurité), nous verrons comment faire évoluer une application Symfony existante tout en restant fidèle à ses principes. L’objectif : montrer que PHP peut écouter, réagir et s’adapter en temps réel — sans renier sa simplicité ni ses patterns historiques.

My session titled “From Black-Box to White-Box: Generative AI Evaluation on Google’s Vertex AI”, will provide attendees with a strategic framework to transition from subjective model testing to rigorous, data-driven evaluation using Vertex AI.

Key Topics will be covered:

Deconstructing LLM Hallucinations: An analysis of the root causes of model errors, specifically statistical auto-completion (“guessing”), fuzzy memory, and the bias explaining why models often prioritize helpfulness over truth.

Vertex AI Evaluation Service: An overview of the fully-managed platform for evaluating Generative AI models and agents, enabling developers to assess criteria like safety, groundedness, and instruction following.

Adaptive Rubrics & Metrics: A deep dive into using adaptive rubrics to generate unique, prompt-specific pass/fail tests that act like unit tests for AI, alongside standard computation-based metrics.

The 3-Step Evaluation Framework: A practical, code-first guide on preparing evaluation datasets, defining and running an EvalTask using the Vertex AI SDK, and visualizing actionable insights via radar charts and Vertex AI Experiments.

Aujourd’hui, on parle beaucoup d’IA dans la tech, mais souvent avec un seul angle. L’idée de cette table ronde, c’est justement d’avoir un regard plus large grâce à trois profils différents, avec chacun un background et une lecture du sujet différents : design et développement front, engineering, et ops / DevOps engineering. Ce sont aussi des profils avec une vraie vision managériale et qui font de la veille, donc l’objectif est de croiser ces regards pour mieux comprendre ce que l’IA change vraiment dans le travail des équipes tech et dans la manière de construire du logiciel.

The ExpressionLanguage component is an old but very useful component. Many well known projects are using it (OroCRM, API Platform, Sylius, Symfony itself, …). This PHP pseudo sandbox is mainly used to create configurations with complex logic.

However, it can also be used as a business rule engine. In that way, applications can be able to let users to do configurations in a dynamic and understandable way. That point is quite interesting. And that’s what I want to cover in this talk.

During the talk, we will outline the ExpressionLanguage component and see its main use cases. Then we’ll take a real-life example and see the component’s potential in an e-commerce context. By following that example, we’ll build together a simple system allowing administrators to write custom discount rules based on customers.

By the end, we’ll be aware of a safe way to delegate some business rules to the user itself.

Le maitre du jeu (MJ) joue un role central dans les jeux de role: il cree les histoires des joueurs, incarne les PNJ et s’assure que les regles soient respectees. L’IA peut etre un formidable allie pour les MJ, en les aidant a verifier les regles, improviser des scenarios et creer du contenu a la volee. Dans ce talk/live coding, nous coderons ensemble un maitre du jeu IA multi-agents a l’aide du SDK open source Strands Agents. Nous mettrons en place des agents specialises pour differentes responsabilites du MJ et verrons comment ils collaborent via le protocole Agent-to-Agent (A2A) et des integrations d’outils grace au protocole MCP.

Avec l’essor de l’IA generative (GenAI), de plus en plus d’applications tirent parti de fonctionnalites liees a l’IA, comme la traduction automatique ou la transcription vocale. Cependant, on peut croire que leur mise en place necessite systematiquement un backend dedie, ce qui peut complexifier le developpement et la maintenance des applications. Ceci ne sera bientot plus le cas grace aux avancees recentes dans les navigateurs web.

Durant cette presentation, nous explorerons les differentes possibilites pour faire de l’inference IA directement dans les navigateurs, en mettant l’accent sur le futur standard Web AI. Nous aborderons les avantages et les inconvenients de chaque approche, ainsi que des exemples concrets d’implementation.

Cette presentation s’adresse aux developpeurs front-end et aux passionnes d’IA souhaitant rendre leurs webapps plus intelligentes. Vous serez surpris de voir ce que peuvent deja accomplir les navigateurs web !

This talk explores how to build and ship intelligent mobile experiences using both remote LLM providers and fully on-device models. We’ll compare Remote, Hybrid, and Local architectures, break down their trade-offs across privacy, cost, and latency, and walk through practical integration patterns using APIs and React Native AI. You’ll leave with clear mental models and production-ready strategies to confidently design, optimize, and scale AI features in real-world React Native apps.

Symfony 8 represents a major leap forward for the framework. As core team members, we’ve helped shape this evolution. As founders of baksla.sh, we’ve lived with the consequences of our architectural choices in production.

This dual perspective—framework development and real-world deployment—has taught us what truly matters when building enterprise applications with Symfony.

In this keynote, we’ll share the patterns and practices that have emerged from this unique position: contributing to the framework while simultaneously pushing it to its limits on business-critical projects.

With Symfony 8.0 and PHP 8.5 both freshly released, the landscape of modern PHP architecture is shifting. We’ll explore what’s now possible and how to make strategic decisions that compound in your favor as your applications grow and evolve.

Join us to discover where Symfony is taking enterprise software development with its latest major version.

Large Language Models (LLMs) are transforming how we build software, but their real business value is still unclear to many teams. In this talk, I’ll share our experience building an AI company on top of this fast-evolving technology — including lessons learned, mistakes made, and what actually worked.

We’ll start by briefly demystifying AI, covering three major categories: statistical AI, generative media (image and video), and language models, with a focus on why LLMs represent a new paradigm for software development.

Next, we’ll explore how to identify meaningful opportunities when concrete use cases don’t yet exist. I’ll introduce practical frameworks we used to separate hype from real value and decide what problems are worth solving.

Finally, I’ll discuss how to communicate a vision that has no direct comparison. We will chat about storytelling techniques we used to help customers, and teammates understand abstract ideas and align around a shared future.

Spreading knowledge and facts in the current era of post-truth, short-form content and deep fakes is more important than ever. History and archives help us deal with the flood of sloppy and misleading information. But working with real archives comes with specific difficulties: large volumes of data, useful information hidden in noise, poor scan quality, handwritten texts that are hard to read.

In this talk, I will explain how we try to solve these problems by using modern technologies while building GRAC LAI, a digital library focused on documents related to the history of Southeast Asia.

We will cover original ways of getting a high-level view of thousands of images, automatic metadata generation for text and images, recent progress in handwritten text recognition models, the use of GPUs in the cloud to speed up processing and the platform to publish and share these archives with the general public.

In a world shaped by accelerating geopolitical tensions and exponential AI adoption, digital infrastructure is no longer a technical choice — it is a strategic one.

Relying on non-sovereign cloud and AI platforms exposes nations and organizations to legal uncertainty, extraterritorial regulations, data dependency, and long-term economic leakage. The cost of inaction is not only financial: it is a loss of autonomy, innovation capacity, and strategic leverage.

Sovereign cloud and AI are not about isolationism. They are about control, resilience, and the ability to decide where data lives, how models are trained, and how value is retained locally. Acting now means investing in trusted infrastructures, local ecosystems, and open technologies — before strategic options quietly disappear.

As web development started in the late 90s, monitors:

• had a 4:3-ratio
• were less than 15-inch wide

INITIAL web design was designed to fit into these devices.

The advent of the smartphone forced the EXISTING INITIAL web design to become “responsive”; the “mobile first” approach was not a new design, but only a new process to achieve the responsiveness of the EXISTING INITIAL web design.

With the advent of large and even over-sized screen displays (large meeting room’s TV, 57-inch 32:9-ratio gamer curved monitors): nothing happened!

No evolution of the existing initial web design was made to fit into these large devices either. But could such an evolution be made? Or is a brand new web design long overdue?

The very simple answer to these questions came from a dissatisfied customer: yes, a brand new design has long been overdue!

The EXISTING INITIAL web design fits into smartphones and monitors with a 4:3-ratio: already on screens with a 16:9-ratio, people are already artificially limiting the width of the page to around 1,200 pixels. And this is exactly the solution provided by CSS libraries such as Tailwind or Bootstrap!

Have you already seen how such websites look like on screens with 21:9 and even 32:9-ratios?

And how would these websites look like in the future on hypothetical Quadri, Penta or HexaFold phones? Already on conventional folded fold phones in landscape orientation, for which the screen ratio is greater than 21:9, these websites have the same design issue.

Now, back to the dissatisfied customer; the word “dissatisfied” I am using about him is a little bit exaggerated: “disappointed” would be more accurate.

After working with him and his own customers on several projects, he entrusted me with the complete redesign of his own company website!

At that time, I was leading the company I had founded, and I had a graphic designer in my team: as a software developer with over 10 years experience in full-stack web development myself, I have the advantage that I can work hand in hand with a graphic designer, check that his/her design is technically implementable, and directly show the website in its demo version to the customer, instead of just a Figma concept.

And, the customer simply reacted: “Michel, what you did looks simply awful!”.

And, he took his phone, and he took a picture of his 57-inch 32:9-ratio gamer curved monitor on which the demo website was displayed, and he sent the picture to me!

And, only at this point I discovered that over-sized screen displays existed.

And, only at this point, I had become aware of exactly the problematic I am exposing in this present paper!

First observation: the “mobile first” approach led to the “one size fits all”, or rather the “one content fits all sizes” mentality. And this cannot be the answer anymore…

ONE answer, mine, is: for ONE range of screen ratios, there shall be ONE specific layout and content:

• screen ratio less than 1:1, ie any device in portrait orientation
• screen ratio greater than or equal to 1:1, but less than or equal to 21:9, ie any device in landscape orientation
• screen ratio greater than 21:9

And, here is the result: https://www.michel-boretti.com/

Second observation: I have been mentioning the different screen RATIOS, but the web development industry is still thinking in FIX SIZES! Including Bootstrap or Tailwind: mr-2, py-4, space-x-1,…

My answer is being made thinking in RATIOS.

More exactly: in RATIO (singular).

Which ONE? Which value?

This is exactly in the presentation I really hope to give on March 2026 in Hanoi!

This presentation introduces Open Infrastructure for the AI Era, highlighting how open-source technologies—led by OpenStack and the OpenInfra ecosystem—are emerging as a foundational platform for next-generation AI and HPC workloads. Drawing insights from the first OpenInfra for AI Working Group white paper, the talk explores reference architectures, best practices, and real-world use cases that demonstrate how open infrastructure can power AI training, inference, GPU-as-a-Service, MLOps platforms, high-performance computing clusters, and AI at the edge.

The session also shares practical experience from FPT Smart Cloud, a Vietnam-based contributor to the global OpenInfra for AI community, including lessons learned from deploying large-scale AI Factory clusters ranked in the Top 500 AI infrastructures worldwide. Key topics include GPU integration strategies (PCI passthrough, vGPU, MIG), high-performance storage and networking (RDMA, InfiniBand, SR-IOV), and system-level optimizations that unlock maximum performance from the same hardware.

By the end of this presentation, attendees will gain a clear understanding of why open, community-driven infrastructure is becoming a strategic enabler for AI innovation—reducing vendor lock-in, accelerating time to market, and creating new opportunities for global collaboration and “Make in Vietnam” AI platforms in the global AI economy.

Most software systems today are built with familiar security practices: trusted administrators, private infrastructure, controlled users, and the ability to fix problems after deployment.

Blockchain systems change many of these assumptions.

This talk shares real examples of what actually changes when current security practices are applied to blockchain-based systems, even when teams follow best practices and have experienced developers.

The first part of the talk is designed to be accessible to non-technical audiences. Through concrete stories from real projects, we’ll explore how ideas like trust, responsibility, ownership, and incident handling evolve when software becomes public, immutable, and permissionless — without diving into technical details.

In the second part, we’ll connect these ideas to real blockchain security incidents, showing how everyday decisions and processes led to unexpected failures once deployed on-chain, and what changes helped avoid repeating the same mistakes.

This is not a marketing talk and not a deep technical session. It’s an honest, experience-based look at how security needs to be rethought when familiar software practices meet a very different execution environment.

AI-powered coding tools have dramatically changed the way we write software. Today, generating large amounts of code is as simple as describing what we want. While Vibe Coding can boost short-term productivity, it also introduces a serious risk: writing and deploying code we do not truly understand — a growing concern for many companies.

AI is a game-changer, and developers must adapt. However, embracing AI does not mean giving up understanding. On the contrary, it offers an opportunity to learn differently and potentially faster.

In this talk, we explore how AI can become a learning partner rather than a blind code generator. We will discuss how developers can utilize AI to gain a deeper understanding of algorithms, language-specific nuances, performance trade-offs, and design decisions. By asking the right questions, code generation becomes a powerful learning tool.

By shifting from Vibe Coding to Vibe Learning, developers can regain control over their codebases, enhance their skills, and build more robust and maintainable software.

AI is no longer a future trend — it’s a present-day career multiplier. This talk explores the current stage of AI, what it can realistically do today, and how professionals can leverage it to boost productivity, expand skill sets, and stay competitive. Attendees will leave with practical insights and actionable ways to elevate their careers in the AI era.

What if AI could become your DevOps pair? In this session, we’ll show how to build an AI-assisted deployment pipeline that understands your code whether you’re deploying a Phoenix app in Elixir, a Rust backend, or a serverless workload. Using LLMs, open-source agents, and CI/CD integrations, we’ll explore how to auto-generate Dockerfiles, Kubernetes manifests, reverse proxy configs, and even sanity tests. This talk is about practical AI tooling for real-world deployments, not just GitHub Copilot magic.

As digital transformation accelerates, organizations must evolve from reactive risk management to intelligence-driven foresight. This forward-looking conference explores how OSINT methodologies, advanced cyber practices, and organizational OPSEC can be integrated into security and compliance functions to anticipate threats before they materialize.

In addition to strengthening resilience, special emphasis will be placed on the protection of critical corporate assets including intellectual property, strategic data, and C-level security against increasingly sophisticated cyber, human, and regulatory threats.

This conference delves into the critical intersection of cyber, human, and compliance risks, highlighting how corporate intelligence acts as a strategic enabler to minimize uncertainty and drive informed decision-making. It explores the essential shift from reactive risk management to proactive, intelligence-driven governance, relying on rigorous cultural foundations and companies transformation.