Many product teams are still in an early, largely ad-hoc phase of AI adoption. Individual PMs may rely on generic chatbots for occasional tasks, but this often remains fragmented and limited in impact. At La Centrale, we took a step back to consider whether meaningful change requires more than simply providing access to AI tools, and whether it calls for a deeper evolution of the Product Management operating model.It’s easy to fall into what we call the “chatbot trap”: using AI mainly for isolated, low-value activities such as summarization, without a shared vision or clear strategy.

‍

We are witnessing a fundamental shift in how users interact with digital products. Static interfaces are giving way to dynamic, conversation-based experiences that mimic human dialogue. Most product teams already build AI agents. The challenge is no longer how to create an agent, but how to make it feel like a coherent, reliable product surface.In this 25-minute talk, I'll share a practical framework to set up AI conversational agents for success, whether they are internal tools or end-user facing products.

Criteo is exploring how AI and agentic capabilities can be integrated into our ad-tech platforms. But how do we decide what should be powered by AI or agents, and how can we ensure that what we build remains useful, coherent, and consistet across all platforms—without slowing down our pace of innovation?

‍

To tackle this challenge, we created the Core AI Design taskforce — a multidisciplinary team bringing together UX Research, UX Design, Content Design, UI Design, and UX Ops. Our mission: to build a robust AI Design Playbook, grounded in research and collaboration, that provides UX Designers, Product Managers, and R&D teams with the tools, principles, and frameworks needed to design AI experiences that truly make sense for our users.In this talk, we’ll share how we built our AI Design Playbook, a framework that bridges human-centered design principles with AI-driven innovation.

We’ll discuss how we evolved from fragmented, product-specific efforts to a cohesive and scalable system that ensures every AI experience—no matter the product or platform—feels consistent, meaningful, and trustworthy.

‍

You rolled out AI tools. Some people love them, some ignore them, and you're drowning in support requests. Sound familiar?Here's what nobody tells you: organizations don't adopt AI, they evolve through it. And each phase needs a completely different playbook.In this talk, I'll share the framework we built navigating three distinct phases, from Ignition through Acceleration to Orchestration, while growing our people.

‍

Anne-Sophie shares Back Market’s 5-year journey building AI capabilities, from a single supply optimization algorithm to 10+ production use cases across supply, finance, and customer care. She covers the key bets behind the results: relentless focus on business impact, build-vs-buy tradeoffs, choosing the right approach (rules, classic ML, or genAI), and stopping underperforming projects. She also explains how their org evolved and how an incubation model helped validate impact before scaling—eventually enabling non-AI teams to run AI POCs autonomously.

‍

‍

‍

AI is reshaping product management: it changes how we prioritize, operate, and design our products. The boundaries between Product Manager, Designer, and Engineer are blurring in favor of profiles capable of experimenting quickly, every day. In this new paradigm, coordination is no longer enough: we need builders with strong research, strategy, and go-to-market muscles. Bpifrance shares the discovery tactics that build confidence in its AI bets—and allow it to invest in the right place, at the right time.

‍

Nelly explains how conversational AI is shifting products from browsing to asking: users state an intent and expect an immediate, contextual response. Drawing on two AI-native products she designed at Hexa—Rose (a marketing agent for B2B websites) and Verso (an audio/video interview agent)—she shows how design decisions and micro-interactions (animations, follow-up prompts, rephrasing, etc.) build trust, boost engagement, and improve the quality of conversations. The talk is hands-on and aimed primarily at designers and PMs.

‍

At Spotify, Guillaume presents Vedder, an internal text-to-SQL tool that turns natural language questions into reliable SQL and helps make data accessible across the company with strong quality guarantees. He shares how the team scaled it from a small experiment to a widely adopted product, improved LLM reliability through continuous evaluation and user-driven data curation, and measured ROI through productivity and adoption—before outlining the path from text-to-SQL to text-to-insights and the question of sustainable differentiation from major LLM providers.

AI has changed product building far more than it has changed engineering.A single founder can now design, prototype, validate, ship, and iterate at a pace that used to require full teams.This talk explains how to build modern products when you’re one person operating in an AI-native environment.

‍

Therry explains how vibe coding makes it possible to go from an idea to a testable product in just a few days, even as new genAI tools appear every day and create confusion. He shares a simple stack—Lovable for the UI, n8n for API interactions and automation, and Supabase for the backend—to ship a fast 0‑to‑1 MVP and validate ideas quickly.

‍

‍

At PayFit, Gaetan and Alexandre will explains how the company is moving beyond one-off AI tools to build an ecosystem of specialized assistants that work together across HR and payroll—for customers, employees, and support agents. He presents a two-pillar strategy: PayFit Copilot for 24/7 autonomous customer support, and an orchestrated ecosystem of CS AI assistants that helps agents handle complexity and focus on empathy and judgment. He also shares PayFit’s 2026 vision of AI shifting from answering questions to taking autonomous actions, with live orchestration examples and lessons from deploying AI in production at scale.

‍

Feedback isn't a tooling problem—it's a culture problem.

In this talk, I'll share a culture-first Product Ops approach to make feedback truly actionable: what teams do (and don't do) to turn raw input into clarity, alignment, and momentum across Product and GTM. Lessons from what I built at Contentsquare, and how we're now scaling it at Gorgias with AI (tagging, clustering, summaries, retrieval). I'm genuinely passionate about this—no magic recipe, just concrete practices you can pick up, adapt, and make your own.

‍

When an organization starts multiplying AI agent use cases, the question is no longer merely technical: it's a matter of governance. Who creates the agents? Who validates the prompts? How can we avoid misuse, black box effects, and security risks? I will share our approach to governing an internal agent platform: responsibility models, validation workflows, quality and risk management."

‍

‍

This talk explores why agents outperform traditional ecommerce UX, how product discovery is shifting from pages to semantic entities, and why structured, enriched product data is becoming the core interface for commerce. We will break down the emerging Agentic Commerce Stack, from catalog foundations and GEO metadata to agent-facing APIs and checkout orchestration, and outline concrete strategies for preparing catalogs and commerce infrastructures for agent-driven discovery and transactions across multiple AI ecosystems.

‍

Martin explains that while AI is making software faster and cheaper to build than ever, speed doesn’t solve product’s hardest challenge: understanding the market and customers well enough to build the right thing. He shares practical strategies to avoid feature bloat and low adoption by improving decision-making—how to prioritize what truly matters, validate customer needs before building, and strengthen human judgment in an AI-driven world.

‍

At Pennylane, the speaker explains how AI is treated as a cultural foundation as much as a business objective, embedded into the company’s core ways of working rather than added as an extra layer to product design. He describes a “virtuous loop” in which internal, everyday use of AI continuously informs the design of AI-powered product features, and vice versa. By building their own AI features, teams gain a deeper technical understanding that makes them more intentional internal users; by relying on AI day to day, they experience the same “aha moments” and frictions as users, building empathy that helps shape more relevant, human-centered product decisions.

Design and deploy agentic systems for the general public on the scale of a French systemic bank. Take into consideration the unprecedented characteristics of a new raw material.

‍

Founder fashion. Excluding Process. Mindset produces radically different products. Pauline will share her double bet:

• combining startup agility and scale-up assets,
• manage the return to constraints without losing momentum.

When building AI features, you can’t anticipate everything upfront. Real user interactions expose unexpected inputs and failures that only appear in production, leaving teams guessing what to fix and whether their changes actually improve the product.

‍

AI coding agents can generate code fast, but most have no idea how your database actually works. They guess column names, ignore security policies, and write SQL that breaks in production. Supabase ships two open-source tools to close that gap: the MCP Server, which gives agents direct access to your Postgres database, and Agent Skills, which teach them to write correct, secure, and efficient SQL. In this live demo, I'll build a working app from a blank project using only Claude Code, plain English, and these two tools.

Many product teams use AI in a piecemeal and fragmented manner, often falling into the “chatbot trap” without a clear strategy. A deeper transformation of Product Management methods is needed. A practical workshop proposes to use an LLM to transform user feedback (categorization of verbatims, identification of problems) into a product prototype.

Michelin discusses the transformation of AI agents through Content Engineering, stressing that content is the driver of these experiences. The objective is to bring out the brand voice in an authentic way by using structured content and clear guidelines to feed AI agents, while focusing on transparency and the value provided to the end user.

LLM applications can subtly fail (cost drift, output degradation, reactive prompt evolution) making production difficult to understand. To ensure the reliability of LLM systems, it is essential to apply practices such as end-to-end monitoring, cost/latency visibility, prompt versioning, evaluations (offline/online), A/B testing, and user data protection. The presentation aims to compare the tools available to measure, debug, and improve an AI product.

Faced with the abundance of AI solutions and the risk of dispersion (the “classic trap” of a “tool-first” approach), the team transformed its experimentation phase into an organized exploration, using its discovery methods to map the “Designer Journey”. This made it possible to identify points of friction with high added value, such as transforming 500 user interviews into insights, in order to move to a “value-first” paradigm.

Many product teams use AI in an ad hoc and unstructured way via generic chatbots, limiting their impact. Real transformation requires a deeper evolution of Product Management methods, in order to avoid falling into the “chatbot trap” (an isolated and low added value use without a clear strategy).

What does it really take to build a “Natively AI” organization and go beyond individual experiments? In this quick 5-minute pitch, you'll dive into the LumApps AI Productivity Program, an initiative designed to encourage the daily use of AI while establishing clear governance throughout the Product Development Cycle (PDLC).

‍