RivieraDev 2025 – Day 2 Highlights

🔑 Keynote: The Cost of Inaction – Quentin Adam & Nicolas Leroux

The second day of RivieraDev opened with an inspiring keynote by Quentin Adam and Nicolas Leroux on a critical theme: “The Cost of Inaction.”

Ken Olsen (1977):
“There is no reason anyone would want a computer in their home.”
A historic underestimation — just like Kodak, who invented the digital camera but didn’t win the market on it, an example of shooting yoursekf in the foot.

Another major example: Blockbuster, which declined a partnership with Netflix — and is now long gone.

📉 Activity vs Purpose

• It’s easy to get stuck in the comfort zone, buried in daily operations.
• Companies like BlackBerry missed the iPhone revolution by clinging to outdated paradigms.

💸 COI vs ROI

• Don’t confuse Cost of Innovation (COI) with Return on Investment (ROI).
• The longer you wait to innovate, the higher the Cost of Inaction.
• Hidden costs include: tech debt, lack of ownership, and stagnation.

Reminder: Not choosing is still a choice — and often the most expensive one.

🌍 Europe vs USA: Two Innovation Philosophies

• The US embraces failure and fast iteration — risk is encouraged.
• Europe tends to favor caution, employment preservation, and regulatory stability.
• But innovation sometimes means letting go — even closing a business if needed.

🚀 Key Takeaways

• Velocity: Tools like ChatGPT, Anthropic Claude, Google Gemini are moving fast.
• Acceleration: We are living the “law of accelerating returns.”
• AI Landscape: The French Vilani report from 2018 was once just used to block a door…
• In software development: visibility matters. Inaction costs you insight, momentum, and relevance.

Personal note: This was a deeply energizing talk — I left the amphitheater with a clear thought: Let’s innovate!

---

🛠️ Martin Kouba – Building Efficient MCP Servers

This session gave us a practical dive into building MCP (Model Context Protocol) servers — and how surprisingly easy it can be.

🧩 What is MCP?

• An open protocol (MIT license) for integrating LLMs with tools and external resources
• Bidirectional communication via JSON-RPC 2.0
• Transports include stdio, HTTP, and (deprecated) HTTP/SSE

MCP provides:

• Server features: Tools, prompts, resources
• Client features: Sampling, roots

⚠️ Missing Pieces in the Ecosystem

• There’s a spec, but no official foundation supports it yet.
• Anthropic maintains the GitHub org, but there’s:
  • No PR review process
  • No OSS governance
  • No shared API standard
  • No TCK (Technology Compatibility Kit) to validate implementations

🎯 Quarkus MCP Module – Goals

• Provide a unified API to implement server-side MCP features
• Expose programmatic access with @Inject

⚙️ Demo with Goose & Gemini

Martin demonstrated how to integrate an MCP server using Goose, a CLI/chat app, connected to Google’s Gemini LLM. The app was reactive, showcasing the real-time nature of:

• Tool usage
• Prompt chaining
• Progress API — for managing long-running tools or resource operations

This session made it clear: while the protocol is still young and lacks formal governance, MCP is promising for building robust AI-enabled tools and agents.

---

🌞 Conclusion – Day 2

This second day was full of energy and meaningful discussions — not only during the talks, but also in the lively exchanges between sessions. That’s part of the true magic of attending RivieraDev.

Let’s not forget the delicious food and local specialties from the South of France: socca, pain bagnat, artisanal ice cream, and of course, the ever-present sunshine!

A fantastic end to an inspiring conference packed with brilliant keynotes, thought-provoking talks, and genuine community spirit. Already looking forward to the next edition!

Tweet

RivieraDEV 2025

RivieraDev 2025 – Let’s Go!

The 2025 edition of RivieraDev is officially underway! Amphitheater 339, with its 500-seat capacity, is completely full. The energy and anticipation in the room set the tone for an exciting lineup of talks and innovations.

---

🎤 Opening Keynote by Thibaut Giraud (aka Monsieur Phi on YouTube)

Talk: “Do LLMs Dream of Electric Knights?”

• LLMs are often referred to as “stochastic parrots” — they generate text without understanding the world.
• Humans also sometimes “parrot” without understanding — language is not always tied to deep comprehension.
• Criticism like “LLMs just predict tokens” is an oversimplification.

Example: Magnus Carlsen’s unpredictable chess moves are strategic and contextual. Predicting them requires deep reasoning — something LLMs are increasingly capable of.

• GPT-3.5-turbo-instruct (as of Sept 2023) can play chess with an Elo rating around 1800.
• LLMs process games encoded in PGN (Portable Game Notation).

Takeaway: Negative results don’t reveal much, it's just teliing you “it failed with this prompt”. Positive outcomes are more telling about LLM potential. It works nows and in the future.

---

🤖 Zineb Bendhiba – MCP in Action

Topic: Integrating AI Agents with Tools via MCP (Model Context Protocol)

• MCP allows local or remote interactions with models (via HTTP, for example).
• LLMs are stateless — apps must manage memory and context.
• LangChain4j supports Java developers building agent-based LLM systems.

---

📚 Guillaume Laforge – LLM Limitations & RAG

Limitations of LLMs:

• No real-time awareness (e.g., current date)
• Can’t access private data
• Hallucinations
• Limited token context (even with Gemini's large windows)

What is RAG (Retrieval-Augmented Generation)?

1. Retrieve: Pull information from external sources (DBs, websites…)
2. Augment: Add retrieved context to the prompt using templates:

   You must answer the question: {{question}}
   Based on this context: {{context}}

3. Generate: LLM produces the final response

Advantages: More accurate, less hallucination, explainable via context, and up-to-date.

Implementing RAG:

• Chunk documents and store in a vector DB
• Use vector search to retrieve and add context

Advanced Techniques:

• Chunking: fixed-length, overlapping, sentence-based, or parent-child structures
• Query compression and routing
• Agentic RAG for multi-step reasoning
• Hybrid RAG + long context models

---

💡 Theo Gianella & Julien Sulpis – How Good Are You at Responsive CSS?

• “Responsive Web Design” coined by Ethan Marcotte (2010)
• Forget pixel-perfect – focus on flexibility and fluidity
• Use modern CSS units: ch, vmin, clamp(), dvh, etc.
• Use container queries instead of relying only on media queries

Reference: CSS Container Query Guide by Ahmad Shadeed

---

🧪 Laurent Dogin – LLM Tools in Nushell

• Nushell + AI agents GitHub repo
• Nushell is a cross-platform modern shell
• Integration with AI uses vector DBs, the ReAct pattern, and TAO prompts (Thoughts, Actions, Observations)
• Key concerns include managing identity, permissions, and chaining tools — similar to BPM pipelines

---

🧭 Arnaud Langlade – Example Mapping (Agile)

Problem: In grooming sessions, devs are passive, meetings run long, and delivery is often misaligned.

“It’s not stakeholder knowledge, but developer ignorance that gets deployed to production.” – Alberto Brandolini

Example Mapping (by Matt Wynne):

• Short (30 min) session with “Three Amigos”: PM, dev, QA
• Use cards or digital tools like Miro:
• Yellow: user story
• Blue: rules
• Green: examples (BDD style)
• Pink: open questions

Tips: Involve the whole team, rotate the scribe, split complex rules, and use examples for tests.

---

🛠️ Dhruv Kumar – Platform Engineering + AI

• Dev productivity is low — only 11% of time is spent coding!
• Cloud complexity, security concerns, and tool overload are major factors

Platform Engineering can help:

• Standardizes SDLC practices
• Security built-in
• Supports tool flexibility
• Improves visibility through DORA metrics
• AI boosts developer experience with smart triage, alerts, and decision support

🎤 Conclusion – Day 1

That’s a wrap for Day 1 of the RivieraDev conference! From thought-provoking keynotes to hands-on tech sessions, the energy and ideas have been nothing short of inspiring.

Let’s see what tomorrow brings as we head into the second — and final — day of RivieraDev. Stay tuned!

Tweet