Dev Summit 2026

This talk will present how we addressed the “dual write” issue we had in our Gifting Order Management solution between our database and our Kafka cluster, causing data inconsistencies and complicated error handling.

The answer ? A simplified version of the Outbox pattern—leveraging a Debezium connector to reliably decouple database writes from Kafka event publishing and using our Order table as the outbox table to simplify the solution.

This architectural shift resulted in a simpler, more robust Order API, eliminating Kafka-specific code and complex error handling. This loosely coupled architecture also offers consumer-driven business logic with easier extensibility.

Join this talk to learn how this architectural decision strengthened our event-driven architecture and discover the crucial details of configuring Debezium—including heartbeat and monitoring—to ensure stability and avoid database disk space issues.

### Introduction:
The key theme is about how product designers can shape the future of an agent powered future for e-commerce. What new challenges do we face and how might we design experiences in this new reality? What new behaviours has this new technology opened up and how can designers use it to provide guardrails?

I want to cover the idea that shopping via AI might sound very cold and transactional, and in the context of Decathlon, that can count against us. Browsing products, becoming emotionally attached to new products and even physical stores themselves is a powerful connection we need to retain. It is a huge differentiator for us.

### The pathways:
Through the lens of three AI powered pathways, I want to explore the opportunities that have opened up for teams at Decathlon and how we might thrive in this new era.

* How do we improve our current experience with new AI tools available on our digital experience?
* What services can Decathlon reference to provide a richer, personalised product experience?
* How do we show up for our customers when they never visit Decathlon in a digital or physical sense?

### Loyalty and emotional connections:
AI use can place an extra layer between Decathlon and our customers, we need to make sure that loyalty is rewarded and is visible when customers use AI to shop.

* How can AI enhance the membership experience, inside and outside of Decathlon's experiences?
* How can we make sure that the repeatable, mundane tasks become automated where possible?
* At the same time, how can we design digital experiences where people do come to us to be inspired, to browse and fall in love with the brand and our products?

The year is 26 after 2000. Digital repositories are fully operational and up-to-date. Well not entirely… Few small application repositories still hold out against some architectural qualities: discoverable, trustworthy, easily consumable, usefulness?
And life is not easy for these repositories: IDPix, LeanIX & Tangramix.

A great Decathlon - ADEO Leroy Merlin FR debate will be organized around these thematix to share best and bad practices:
- « Aren’t my data fresh? » : how we continuously insure the data freshness of our repositories?
- « I’m not fat, I'm just big boned! »: how do we add new data in our repositories and discard deprecated ones?
- « These Repos are crazy! » : how do we promote IT use cases on top of these repositories (Dev, Architecture, FinOps, IA…)?

Text by Ahmed, Fabien, Gauthier and Laurent & Drawing by Uderzo.

# Abstract:
LLMs (Large Language Models) like ChatGPT have entered the industry with a velocity that no retailer can ignore. In just a few years, usage has exploded to over 1 billion weekly users, with record amounts of capital pouring into the sector.
This represents a fundamental shift in how people interact with digital content—a revolution comparable to the advent of the Internet itself. A significant portion of human digital attention has shifted to this new medium, and business models are following.
We will discuss the different phases of this evolution and what it means for our brands Adeo and Decathlon. Where we think we are. How we believe it is about winning a war of attention in a new sales channels and what we are doing about it. We will also explore the potential risks that we have to mitigate along the way.

# Plan:
- Intro with the different phases of AI's impact on E-commerce. Look at the past to understand the future.
- Winning the attention battle - Why it matters. Some examples for our 2 brands (examples: GEO initiatives, ChatGPT App, POC with Mirakl, ....)
- Enriching our product sheets to promote discovery by LLMs (putting our product pages into perspective and their evolutions related to AI)
- AI Content and UGC
- Risk Analysis (we often don't talk about it enough, LLMs are a formidable opportunity but the path is filled with risks from brand and reputation to cost ).

# Bonus :
The last section is focused on using & benchmarking multiple AI Models (Claude, ChatGPT, Gemini, Kimi) to **imagine & design joint/common Product / Services / Tech Capabilities between Adeo & Decathlon in the future**

Four value streams, three and a half developers, all moving simultaneously. There was pressure to start everything — it looked like progress. But at the end of each sprint, delivered value was scarce. The data told a different story.

We ran an AI-assisted analysis of our historical flow metrics: WIP over time, lead time per story point, where epics actually got stuck. The pattern was clear — we had a systemic problem, not a capacity one. Starting earlier was making us finish later.

The fix was counterintuitive. We reduced the number of topics in flight, moved from individual domain ownership to team-based delivery, and realigned our backlog around collective throughput. The analysis showed a potential 2x improvement in lead time by reducing WIP — and early signals are promising.

We'll walk through the data that changed our thinking, the steps we took, what didn't go as planned, and what surprised us. Fair warning: the experiment is still running — no clean happy ending, just honest numbers and the reasoning behind every decision.

What you'll take away: a replicable approach to focused, team-based flow — and how to make "let's finish fewer things faster" a convincing argument.

In a marketplace with 17M+ products and 42M+ offers, every unpublished offer is lost revenue. But why is an offer not live? The answer used to be a mystery trapped in technical logs.
To solve this at scale, we didn’t just build a dashboard; we built a framework for clarity.
Join us behind our new Offer Management system to learn how we used the Pyramid of Failure to categorize user, business, and technical blockers. We’ll share the blueprint for building a single, scalable UI that serves diverse personas, empowering sellers with autonomy while providing tech teams one "Source of Truth" for debugging and investigation, and how this unified approach turned technical friction into GMV growth.

Leroy Merlin Italia is completing a crucial phase of its digital transformation. Over the past years, we've built new platforms, adopted new tools, redesigned architectures, and invested heavily in digital capabilities. Along the way, we've collected powerful lessons learned and now we're putting them to work.

The next step is clear: **simplification and productivity**. We need to train our teams for process, not for product. We need to complete the human and mindset transformation, steering our squads toward cross-functional thinking and process efficiency, subtracting the superfluous instead of adding more. We need to leverage AI both to streamline business processes and to accelerate our tech and digital delivery (i.e. MVPs, prototypes, requirements, user stories, test cases) to unlock the potential of the solutions we've already built and to accelerate business outcomes.

In this post-onboarding phase, the real value that a digital team can bring to the organization is something that no single business direction can provide on its own: **a cross-functional, end-to-end vision of processes.** Business directions are by design and by vocation, focused on their vertical. That's their strength. But the digital team sits at the intersection of all verticals, and that unique position makes it the natural driver of process simplification, waste elimination, and horizontal efficiency across the entire value chain.

But here's what years of transformation taught us: **if you automate a mess, you just get a faster mess.** And nobody tells you that.

The Digital Lean Framework (DLF) is a framework that I conceived and designed with the support of my team, as a natural evolution of our purpose at Leroy Merlin Italia. It takes the core principles of lean thinking, born on Toyota's factory floors 70 years ago and applies them to digital, tech, cybersecurity, and organizational development. Not as theory, but as a proprietary operational method distilled from 20+ years of enterprise delivery and shaped by the lessons we've learned building, failing, and rebuilding in the field.

In this talk, I'll share:

- **Legge Zero**: why starting from technology to shape the process (instead of the opposite) is the original sin of digital transformation, and how to reverse it.
- **The 8 Digital Muda**: the invisible wastes killing your teams' productivity, from overprocessing and waiting to the most dangerous one: the **Perfection Trap**, a waste that doesn't exist in manufacturing but is devastating in digital.
- **A practical framework**: (VALUE → STREAM → CLEAN → FLOW → KAIZEN) to find and eliminate waste in any digital process from IT service delivery to product development.
- **Real results** from applying a first dry run of DLM within a bunch of use cases at Leroy Merlin Italia: what we found, what we cut, what improved, what assumptions we confirmed, what assumptions we destroyed, and what we're still learning.

This is not a talk about doing less. It's about doing what matters, with method, data, and relentless focus on value.

*"If you automate a mess, you just get a faster mess." — Digital Lean Framework*

In 2026, "just chatting" with an AI isn't enough. We need agents that can act: query a database, trigger a deployment, or check real-time stock levels. Enter the Model Context Protocol (MCP)—the universal standard for connecting AI models to local and remote data sources.

This hands-on lab is a deep dive into building the "connective tissue" of modern AI. Using Spring AI, we will move beyond basic API calls to develop robust MCP Servers in Java. You will learn how to expose your existing business logic as "tools" that any LLM can discover and use securely. By the end of this session, you’ll know how to transform your Java applications into specialized plugins that give your AI the "hands" it needs to interact with your enterprise systems.

Depending on a single vendor for all data needs creates significant bottlenecks: vendor lock-in, unoptimized compute engines for specific workloads, and slow developer feedback loops. The next phase of data architecture demands a shift toward decoupled layers and flexible compute.

In this session, we will explore the future of data transformation built around open table formats and multi-engine processing. We will discuss how open formats like Apache Iceberg can turn Data Mesh theories into reality, while addressing current catalog challenges and where the ecosystem is heading next.

Finally, we will focus on enhancing the developer experience through local compute. We will run a practical demonstration where a Databricks/BigQuery dbt project is executed locally on DuckDB using automated SQL dialect translation. The result: zero cloud compute costs during development and lightning-fast iteration cycles for your data engineering teams.

We are excited to present our Copilot-accelerated migration of the Last Mile product from Scala to Java. Last Mile is a critical tool for Leroy Merlin France, designed to streamline the sales, management, and tracking of over 1.2 million deliveries every year. By ensuring operational excellence, it acts as a primary driver of our customer satisfaction. We want to talk about the project's context, our core goals, the step-by-step migration process, and our key learnings.

This 90-minute Hands-on Lab invites you to discover a new approach by leveraging the power of the Release Management. The goal is simple: turn every deployment into a "non-event"—perfectly predictable, repeatable, and secure.

On the agenda (100% hands-on):

- Express Onboarding: Get hands-on with the platform and onboard your digital product in minutes.
- Standardization & Promotion: Create your product releases and use promotion pipeline templates to ensure consistent reliability across all environments.
- Guardrails & Robustness: Use automated checks and compliance policies to block errors before they ever reach production.
- Visibility & Governance: Track and manage your deployments via a unified dashboard for total control.

The result? You'll be autonomous to transform your digital product in the new culture of the Release Management!

Ready to deploy your next release Friday evening?

Most developers know **Valkey** as a high-performance distributed cache, but how many truly leverage its potential as a versatile data structure server? If you've only ever used `GET` and `SET`, you are missing out on the features that make Valkey a Swiss Army knife for modern architectures.

In this codelab, we will strip away the abstractions and go back to basics. Using a practical, step-by-step approach, you will interact with Valkey through its CLI and understand its core mechanisms.

**Pre-requisite:** come with Docker installed

**Note:** Valkey is an open-source, high-performance key-value store that was created as a fork of Redis (specifically version 7.2.4) following its transition to a non-open-source licensing model.

What happens when a customer checks the status of an order and finds different answers depending on the channel?
This question was the starting point of an investigation that revealed a bigger issue than it initially seemed: the same order could display different information in the app and on the web, regardless of whether it was placed through e-commerce or in a physical store. The outcome was predictable — customer confusion, increased Contact Rate, and a direct impact on NPS.

To understand the root cause, we initiated a **discovery process** that mapped the entire order journey across multiple customer touchpoints — app, web, telesales, email, and WhatsApp — while also diving into the systems that generate and distribute this information.
It quickly became clear that the challenge was not only technological, but about **orchestrating systems, channels, and teams to transform a complex ecosystem into a simple and reliable experience for the customer**.

Collaboration between squads enabled us to standardize information in the backend, increase transparency for the frontend, and evolve the order tracking experience — making it easier for customers to check order status, refund information, and in-store pickup details.

In the end, the goal is as simple as it is powerful: ensuring that customers can find a **single source of truth about their order**, no matter where they look. As a result, we aim to reduce Contact Rate, increase the app’s NPS, and decrease the number of abandoned payments.

In this talk, we will share the behind-the-scenes of this discovery process — and how technical and product decisions can directly transform the customer experience.

🚀 If the word **"SAP"** still triggers flashbacks of grey screens, clunky transaction codes, and endless PDF documentation... it’s time for an intervention. 🛑

Join us for a **30-minute deep dive** into the ultimate developer glow-up. We’re moving beyond the "legacy" mindset to show you how LLMs and the Model Context Protocol (MCP) are fundamentally rewriting the rules of the game.

In this session, we’ll take a deep dive into this transformation of the developer experience:

🤖 How an SAP developer can go from a natural language prompt **to ABAP/RAP** code that is generated, validated, and deployed directly to their system — all powered by AI.
🧠 We'll cover how **LLMs work** and how to leverage them effectively in an SAP context, prompting strategies and the art of orchestrating RAP and Fiori code generation, the **Model Context Protocol (MCP) ** and how it bridges AI with your SAP tooling — **ADT, documentation, the Cloudification Repository**.

🌍 **Who is this for?**
Whether you’re an **ABAP dev or a dev** curious about the "New SAP," come see how we’re building the future today.

No more "tomorrow"—the AI-powered ERP is already here.

This 90-minute Hands-on Lab invites you to discover a new approach by leveraging the power of the Release Management. The goal is simple: turn every deployment into a "non-event"—perfectly predictable, repeatable, and secure.

On the agenda (100% hands-on):

- Express Onboarding: Get hands-on with the platform and onboard your digital product in minutes.
- Standardization & Promotion: Create your product releases and use promotion pipeline templates to ensure consistent reliability across all environments.
- Guardrails & Robustness: Use automated checks and compliance policies to block errors before they ever reach production.
- Visibility & Governance: Track and manage your deployments via a unified dashboard for total control.

The result? You'll be autonomous to transform your digital product in the new culture of the Release Management!

Ready to deploy your next release Friday evening?

You have spent years designing robust architectures, optimizing system performance, and automating everything that can be. As an Individual Contributor (IC), your impact is clear, measurable, and direct. But there comes a time when, to truly move the needle, you need to change your paradigm: what if your next scalability challenge wasn't technical, but human?

Moving from a technical expert role to a manager role should not be seen as a "promotion" up the hierarchy, but as a transition to a completely new profession where your success depends on the success of others. This is the story of moving from individual optimization to building a system capable of growing without you being a Single Point of Failure (SPOF).

Through this feedback from a former architect and open-source enthusiast turned Engineering Manager, we will explore the stages of this mutation:

* **Daily Impact:** How to redefine your "Definition of Done" when you no longer produce code?
* **Becoming Scalable:** Learning to delegate not to offload work, but to multiply the organization's impact.
* **Technical Legitimacy:** How to remain a credible peer without becoming your team's bottleneck.
* **Beginner Pitfalls:** Identifying common "bugs" for new managers (micromanagement, keyboard withdrawal, imposter syndrome).

This talk is for anyone questioning their future career path: should I remain an expert or switch to management? The goal is to provide concrete keys to successfully making the leap, understanding that management is ultimately about designing an architecture where humans are the most precious and complex components.

**Keywords:** Leadership, Engineering Management, Career Path, Transition, Scalability, Soft Skills, Feedback, Coaching.

Discover the Discovery to Delivery process in the End user computing perimeter in Decathlon, using a mixed Jira Discovery and Jira Software to follow the delivery process of an idea !

Since 2021, in the End User Computing team, which covers mainly hardware management in Decathlon, we tried many different methods to adopt and improve our Discovery in our Product Manager routines.

Now participating in the Product Guild : Tools, we formalized a minimalist stable process and participated in the creation of a Jira Discovery template that can be easily adopted by beginners and seniors in their Product Management journey.

In 2026, "just chatting" with an AI isn't enough. We need agents that can act: query a database, trigger a deployment, or check real-time stock levels. Enter the Model Context Protocol (MCP)—the universal standard for connecting AI models to local and remote data sources.

This hands-on lab is a deep dive into building the "connective tissue" of modern AI. Using Spring AI, we will move beyond basic API calls to develop robust MCP Servers in Java. You will learn how to expose your existing business logic as "tools" that any LLM can discover and use securely. By the end of this session, you’ll know how to transform your Java applications into specialized plugins that give your AI the "hands" it needs to interact with your enterprise systems.

The talk aims to explain why NPS does not tell everything about an e-commerce experience. Satisfaction scores can look healthy while key journeys remain inefficient. The focus then shifts to fixing the basics: essential UX tactics and minimal features required to meet client expectations and reach market standards.

Building a product configurator at Adeo has historically been a months-long endeavor: modeling compatibility rules, translating business constraints into code, endless test cycles… And evolving an existing rule? Days of work for the tech team.
With our CPQ (Configure, Price, Quote) project, **winner of the Adeo XP Awards 2026 in the Disruptive Innovation category**, we flipped this paradigm entirely. The only skill you need to use CPQ is the ability to write in your own language. A business expert describes a rule in plain words *"if the customer picks a floorboard wider than 15cm, only show compatible underlays"* and Gemini automatically generates the execution logic, along with a plain-language explanation to review before going live.
What used to take weeks now takes minutes. In this talk, we'll share how we built this AI pipeline (Spring AI + Vertex AI), the challenges we faced, and the lessons learned from a project where AI isn't a feature — it's the foundation.

I work as AI Governance Lead at Decathlon with a backgrouns in responsible AI and AI safety research engineering. In this talk, I'll present a paper that I've published with NeurIPS and AAAI "Empirical Evidence for Alignment Faking in a Small LLM and Prompt-Based Mitigation Techniques". I have presented a similar talk at the AAAI Fall Symposium Series last year. Given the audience at this summit, I can also spend some time diving into the importance of AI safety in multinational organisations and how we can go beyond policy, to include technical AI safety measures.