Observabilité des processus
A local-first, privacy-focused markdown editor. No accounts, no cloud dependency, no sync fees — just a folder of markdown files you fully own.
A local-first, privacy-focused WYSIWYG Markdown vault with full-text search, wiki-links, and a rich editor. Built with Tauri, Svelte, and Rust
Reduce noise in decision making
Plought separates setup from evaluation so you can weigh priorities, work through the tools, and review the tradeoffs before you choose an option.
TODO
- fork pour supprimer l'IA
A modular, local-first block editor engine.
Single contenteditable. Loro-CRDT.
Free, fast, and offline PDF toolkit for professionals. Edit, convert, and process PDFs entirely in your browser with no data uploads. 100% client-side processing.
The professional PDF toolkit that runs entirely in your browser. Powerful WASM engine, no server uploads, simply secure.
Une interface pour configurer Ghostty avec une apparence macOS
Je suis développeuse web depuis une grosse quinzaine d’années. Les IA génératives sont en train de me dégoûter de mon métier.
Je déteste avoir le ventre qui se tord et la voix qui se brise quand j’essaie d’expliquer à quel point je les déteste. Alors j’écris, les yeux brillants.
Je déteste l’idée d’être dans cette industrie qui, globalement, continue d’utiliser ces outils. Chaque problème qu’ils posent, pris séparément, suffit à mes yeux pour s’en passer, et pourtant mes pairs continuent de les produire, de les défendre, de les utiliser. J’en ai marre de ne plus pouvoir faire trois pas dans le monde de la tech sans croiser un gars qui m’explique que tous les problèmes de l’IA sont causés par les utilisateurs qui ne savent pas s’en servir correctement.
Je suis dégoûtée.
C’est assez nouveau, en fait. Pourtant, je suis une femme qui a étudié en école d’ingénieurs et travaillé dans diverses boîtes d’informatique pendant seize ans : des dégoûtants, j’en ai côtoyés. Des machos, des violents, des alcoolisés, des vieux et des jeunes, avec et sans cravate. Curieux, en fait, que je n’aie pas été dégoûtée plus tôt, par ces dégoûtants tellement mieux payés que moi, comme tant d’autres femmes de la tech avant moi.
Mais les IA génératives, c’est différent.
Elles détruisent tout. L’environnement. L’humanité. Et, c’est là que ça devient personnel, elles détruisent très précisément ce que j’aime dans mon métier pour mieux me noyer dans le reste.
Personnellement je refuse de me servir des IA génératives ; je mesure la chance d’être mon propre employeur et d’avoir la liberté de refuser. Tant d’autres ont expliqué ce qui ne va pas avec les IA génératives. Je ne ferais que répéter. Je vais essayer de me concentrer ici sur ce qui m’a fait aimer ce métier.
De ce métier, et dans l’absolu, j’aime principalement trois choses : créer, apprendre et transmettre. Le métier de dev a ceci de sympathique — en tout cas il avait ceci de sympathique avant 2022 — qu’il permet de varier les plaisirs en permanence avec des petites combinaisons : apprendre en créant, créer pour transmettre, et apprendre en transmettant.
Peut-on battre les modèles de Google ou Meta avec seulement 4 GPU et une disquette Zip ? C’est le pari fou de notre invité.e qui nous explique comment le "Data Design" est en train de ringardiser le scraping massif du web. 🥖 L'IA qui tient sur une disquette : La fin du gigantisme ? Dans cet épisode, on plonge dans le coeur de l'IA souveraine : pourquoi la qualité des données (tokens) prime sur la quantité, et comment les Small Language Models (SLM) vont permettre de décentraliser l'intelligence. 🚀 Ce que vous allez apprendre :
- Baguettotron : Le modèle de 320M de paramètres qui raisonne mieux que des géants.
- Data Design vs Scraping : Pourquoi "nettoyer" la donnée ne suffit plus, il faut la concevoir.
- Le secret des données synthétiques : Comment éviter le "Model Collapse" (l'appauvrissement de l'IA).
- Souveraineté : L'enjeu des bibliothèques nationales et de l'Open Data face au pillage des "Shadow Libraries".
⏳ Timestamps pour naviguer : 00:00 — Jeu d'indices : qui est la pionnière de la tech française ? 04:38 — L'arnaque du "poids ouvert" : qu'est-ce qu'une IA vraiment Open Source ? 14:41 — Data Design : pourquoi Pleias mise sur la provenance plutôt que le scraping 24:11 — Baguettotron : l'IA performante qui tient sur une disquette Zip 36:01 — Small Language Models (SLM) : battre les géants avec seulement 4 GPU 52:00 — L'avenir décentralisé : IA locale, souveraineté et modèles de raisonnement SPOILER ALERT : pour en savoir plus sur notre invitée Anastasia Stasenko , CEO Pleias : https://www.linkedin.c... 🔗 Liens et ressources : Pleias : https://pleias.fr/ Modèles & Datasets : Retrouvez "Common Corpus" sur Hugging Face.
Christopher Alexander’s keynote speech at the ACM Convention on “Object Oriented Programs, Systems, Languages and Applications”, (OOPSLA) in San Jose, CA, October 6 - 10 1996. Christopher Alexander connects his architectural work on patterns and pattern languages [03:40] with the field of computing. He introduces the theoretical framework behind his upcoming work, "The Nature of Order" [04:12], and expresses concern about the lack of "living structure" in the modern built environment [05:52]. Alexander emphasizes the moral component underlying his work, questioning if computing patterns also strive to improve human life [13:39]. He discusses the importance of a pattern language's ability to generate coherent wholes [17:32], the recursive structural characteristics related to patterns [21:11], finding objective criteria for "life" in structures [24:57], and the concept of "unfolding wholeness" [41:25]. He concludes by urging the computing community to consider their potential role in creating living structures in the world [59:22].
Transcript : https://www.patternlanguage.com/archive/ieee.html
From Technical Debt to Cognitive and Intent Debt:
Rethinking Software Health in the Age of AI
Margaret-Anne Storey, University of Victoria, Canada
March 23, 2026
Generative AI is dramatically accelerating the velocity of software development, enabling small teams to ship
features at a pace that would have seemed implausible just a few years ago [Peng et al. 2023]. I saw this
firsthand in an entrepreneurship course I taught recently. Student teams were building software products over the
semester, moving quickly to ship features and meet milestones. By week eight, one team hit a wall. Simple
changes were breaking things in unexpected places, and progress had stalled. When I met with them, they
initially blamed technical debt: messy code, hurried implementations, architectural shortcuts. But as we dug
deeper, a different problem emerged. No one on the team could explain why certain design decisions had been
made, or how different parts of the system were supposed to work together. The code might have been messy,
but the deeper issue was that the team's shared understanding, the theory of the system [Naur 1985], had quietly
fragmented. They had also failed to write down or communicate the rationale behind decisions. They had
accumulated cognitive and intent debt faster than technical debt, and it had paralyzed them.
This is not an isolated story [Willison 2026]. Generative AI does not remove the challenges of software
engineering; it redistributes them. In this article, I propose a triple debt model for reasoning about software
health, built around three interacting debt types: technical debt refers to problems in the code layer, cognitive
debt refers to inadequate understanding across a team, and intent debt refers to a lack of externalized rationale,
information that both humans and AI systems need to work safely and efficiently with the code. Technical debt
makes systems harder to change. Cognitive debt makes systems harder to understand. Intent debt makes it
difficult to know what the system is actually for.
A few decades ago, the only well-known way to deliver something to a server, to make it accessible over the internet, was moving files via FTP in Total Commander, FileZilla or FAR Manager, manually copying files and folders from the left pane to the right one. The more advanced among us preferred standard UNIX tools like scp or rsync instead, but the process was essentially the same.
...
The quiet revolution happened in 2000. Not on Windows Server, and not yet on Linux — but on FreeBSD, a UNIX-based operating system that was the default choice for IT professionals long before Linux dominated the space.
FreeBSD is worth a brief aside here, because it differs from Linux in a fundamental way. Linux is a kernel. What most people call "Linux" is actually that kernel combined with a GNU userland, a package ecosystem, and a set of choices that vary from distro to distro — Ubuntu, Fedora, and Arch are all running the same kernel but are meaningfully different systems underneath.
FreeBSD ships as a complete, coherent OS — kernel, userland, base tools, and libraries all developed together, versioned together, and tested together as a single unit. That coherence matters. It's part of why FreeBSD solutions tend to be cleaner and why the base system behaves consistently across installations.
The solution FreeBSD built on top of that coherent foundation was called jails. Announced by Poul-Henning Kamp and Robert Watson and shipped as a native kernel feature in FreeBSD 4.0 in March 2000, jails took the chroot idea and completed it — adding full network isolation, process isolation, and proper security boundaries.
...
Design patterns promise reusable solutions to recurring problems, yet many remain unused, misunderstood, or applied mechanically. This paper argues that patterns often fail because they insufficiently account for the umwelt—the situated perceptions, experiences, intentions, and values of authors, designers, users, and the contexts in which patterns are enacted. Building on Christopher Alexander’s notion of living structure, we reconceptualize patterns as potentials rather than finished solutions: seeds that must be unfolded through situated design decisions. We identify common “pattern smells” that hinder effective use and show how attention to unfolding, roughness, and contextual fit can make patterns more resilient and meaningful. Extending this perspective, we emphasize the importance of vividness in pattern descriptions—clarity, conciseness, and expressive forms of representation that engage designers, invite reflection, and support learning across diverse umwelts. We conclude with guidance for writing, maintaining, and evolving patterns so that both their descriptions and implementations remain alive, adaptable, and capable of resonating with changing environments and practices
Open Buro
The open European standard for workplace orchestration.
Open Buro is the missing layer that turns isolated open source workplace apps into a unified digital platform — capable of rivaling Microsoft 365 and Google Workspace, without vendor lock-in.
The Problem
Europe has mature open source alternatives for every workplace function: email, documents, project management, video conferencing, chat, calendar. But these tools remain isolated silos. Even with SSO, users get an app catalog — not a platform.
Meanwhile, dominant suites win not because they're better app-by-app, but because they deliver a platform effect: an interlocked ecosystem where everything flows together effortlessly. Leaving becomes perceived as risk, not a project. The dependency is strategic and political, not just budgetary.
The Standard
Open Buro defines how independent open source services assemble and communicate through a common orchestration layer. The standard covers 7 domains:
- Application Integration — Unified SSO, standard app packaging, centralized registry, common settings API
- Cross-service Navigation — Shared home screen, unified nav, global app grid, cross-app command palette
- Data Intelligence — Business object definitions, cross-app event streaming, knowledge graph, unified search
- Platform Collaboration — Cross-service workspaces, threaded comments, unified notifications, shared presence
- Inter-apps & AI — Capability/intent casting, shared file picker, AI agents orchestrating across tools
- Security & Encryption — Platform-level E2E encryption, granular permissions, audit logging
- Mobile & Desktop — Native mobile apps, desktop client, browser extension
The Alliance
The Open Buro Alliance is the collective movement — publishers, institutions, governments — that governs and promotes the standard. Structured as a neutral foundation (modeled on Linux Foundation / CNCF), no single vendor controls the standard.
Founding members:
- Twake (LINAGORA's collaborative platform)
- La Suite numérique (DINUM, French government digital workplace)
We are at the very beginning, join us !
