MACI #147 - Rust, Materia, Open Source and my thoughts on AI

In this episode of Messages à caractère informatique (#147), I focus on two key themes that matter in today’s tech landscape: how we design connected systems and how we use AI efficiently as developers.

On the cloud side, I react to the story of cloud-dependent smart beds that stopped working when their provider went down. My angle is pure UX design and system resilience: if a “default mode” can trap people in an uncomfortable position all night because a DNS or cloud service fails, the issue isn’t AWS or the cloud itself, it’s poor product design. I highlight a fundamental principle of reliable connected devices: you must always provide a local, manual fallback for smart beds, garage doors or fully automated homes during power outages.

In the episode, I also introduce Fork it!, the international tech conference series I co-organize. Fork it! aims to connect developers and communities worldwide, essentially building a human neural network that encourages knowledge-sharing across ecosystems, cultures and technologies. It’s a way to strengthen the human side of tech in a world dominated by automation and AI.

When the discussion moves to AI tools and developer workflows, I explain that many developers use AI backwards: they try to automate the part where they actually bring value: writing code, using models they don’t fully understand. I take the opposite approach. I use AI to:

• grasp concepts I don’t fully master yet (LLMs, AlphaGo/Katago-style systems, “AI-compatible cloud” buzzwords),

• write and refine specifications and tickets, instead of dropping vague, incomplete issues,

• improve documentation, clarity and structure: the essential but often neglected parts of a project.

Meanwhile, the others explore topics like SQLite reimplemented in Rust with vector search for LLMs (Turso), embedded and IoT database use cases, the Tarmageddon Rust tar vulnerability, and the broader issue of open-source maintenance. They also discuss GitLab Knowledge Graph with AI and MCP to map dependencies, Google’s latest robotics and Gemini models, the constant FOMO around new AI tools, and AltStore PAL for iOS sideloading under the EU DMA.

Throughout the episode, I bring a practical perspective focused on UX, safety, resilience, and using AI where it actually improves your workflow instead of chasing every new hype tool.