<script data-pm-proxy="intercept"></script><?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:itunes="http://www.itunes.com/dtds/podcast-1.0.dtd" xmlns:googleplay="http://www.google.com/schemas/play-podcasts/1.0"><channel><title><![CDATA[Veille de la semaine]]></title><description><![CDATA[Un récapitulatif hebdomadaire des articles / tweet / sujets marquants]]></description><link>https://guikingone.substack.com</link><image><url>https://substackcdn.com/image/fetch/$s_!as1P!,w_256,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fguikingone.substack.com%2Fimg%2Fsubstack.png</url><title>Veille de la semaine</title><link>https://guikingone.substack.com</link></image><generator>Substack</generator><lastBuildDate>Thu, 03 Sep 2026 09:11:42 GMT</lastBuildDate><atom:link href="/__u/guikingone.substack.com/feed" rel="self" type="application/rss+xml"/><copyright><![CDATA[Guillaume Loulier]]></copyright><language><![CDATA[en]]></language><webMaster><![CDATA[guikingone@substack.com]]></webMaster><itunes:owner><itunes:email><![CDATA[guikingone@substack.com]]></itunes:email><itunes:name><![CDATA[Guillaume Loulier]]></itunes:name></itunes:owner><itunes:author><![CDATA[Guillaume Loulier]]></itunes:author><googleplay:owner><![CDATA[guikingone@substack.com]]></googleplay:owner><googleplay:email><![CDATA[guikingone@substack.com]]></googleplay:email><googleplay:author><![CDATA[Guillaume Loulier]]></googleplay:author><itunes:block><![CDATA[Yes]]></itunes:block><item><title><![CDATA[Veille de la semaine du 24 Août 2026]]></title><description><![CDATA[Simuler la nature en Javascript, du backup pour vos bases de donn&#233;es, basculer sur Rust en venant de Go, une backdoor dans vos LLMs et PHP qui se met au C++, pr&#234;t(e)s ?]]></description><link>https://guikingone.substack.com/p/veille-de-la-semaine-du-24-aout-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/veille-de-la-semaine-du-24-aout-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 28 Aug 2026 14:01:56 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/213058104/899d32822eb5baf0e4c25cf0318956a3.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>B</strong>ien le bonjour &#224; tous et &#224; toutes, j&#8217;esp&#232;re que tout le monde va bien ?</p><p>La Veille de la semaine arrive comme &#224; son habitude en ce nuageux vendredi, au programme ? De la culture g&#233;n&#233;rale, un peu de PHP, du Machine learning et le retour de la base de donn&#233;es, le tout, servi avec quelques avis tranch&#233;s et toujours sa dose de podcasts.</p><p>Bref, parlons peu, parlons bien, on se met &#224; table ?</p><div><hr></div><h1><strong>Culture g&#233;n&#233;rale</strong></h1><p><strong>O</strong>n commence par la culture g&#233;n&#233;rale et notamment la d&#233;couverte de Nature of Code, un livre expliquant comment r&#233;pliquer les syst&#232;mes naturels via Javascript, j&#8217;avoue ne m&#8217;&#234;tre jamais pench&#233; sur certains sujets (j&#8217;utilise des vecteurs et des r&#233;seaux neuronaux au quotidien mais de l&#224; &#224; en reconstruire un, un monde existe ...) et voir du code qui reste &#8220;lisible&#8221; et peu soumis &#224; des d&#233;pendances externes fait plaisir.</p><p>Autre sujet avec une tr&#232;s courte vid&#233;o sur la construction d&#8217;une console permettant de lancer la s&#233;rie Halo dans un format compact et &#8220;sur-mesure&#8221;, le visuel est diablement efficace, le format passe partout, bref, &#224; surveiller.</p><ul><li><p><a href="https://natureofcode.com/">Nature of Code, simulating natural systems in Javascript</a></p></li><li><p><a href="/__u/beefandbitcoin.substack.com/p/pourquoi-sommes-nous-autant-soumis">Why are we enslaved to the State? Testosterone might be an answer</a> - French</p></li><li><p><a href="https://www.youtube.com/watch?v=1RJcOIpp2OQ">Building the console Microsoft never will</a> - Youtube</p></li></ul><div><hr></div><h1><strong>Base de donn&#233;es</strong></h1><p><strong>O</strong>n continue avec la base de donn&#233;es et notamment Databasus, un outil self-hosted permettant d&#8217;effectuer vos backups sur vos bases de donn&#233;es favorites (MySQL, MariaDB et d&#8217;autres sont &#233;galement support&#233;es), le tout en b&#233;n&#233;ficiant d&#8217;une UI sobre mais efficace, la possibilit&#233; de planifier vos backups, de les stocker hors de la machine en cours, bref, du tr&#232;s s&#233;rieux et disponible &#233;galement via Docker (plus d&#8217;excuses).</p><ul><li><p><a href="https://github.com/databasus/databasus">Databasus, an open-source and self-hosted tool to backup PostgreSQL</a> - Github</p></li><li><p><a href="https://github.com/jeffhajewski/latticedb">LatticeDB, a property-graph database</a> - Github</p></li></ul><div><hr></div><h1><strong>Cloud computing / Linux</strong></h1><p><strong>T</strong>out petit &#233;cart du c&#244;t&#233; de Linux pour un article (qui aurait pu finir dans la culture g&#233;n&#233;rale au passage) revenant sur le passage de Go vers Rust et les embuches rencontr&#233;es en cours de route, &#233;tant un d&#233;fenseur de Rust (et n&#8217;appr&#233;ciant que peu Go, je le confesse), je dois dire que les arguments se tiennent et sont souvent ceux qui remontent quand on parle de Rust et d&#8217;une &#8220;bascule&#8221; pour des outils en CLI et autres joyeuset&#233;s.</p><ul><li><p><a href="https://github.com/djc/oxish">Oxish, a Rust-base memory-safe SSH server</a> - Github</p></li><li><p><a href="https://rolandsdev.blog/posts/from-go-to-rust/">From Go to Rust</a></p></li></ul><div><hr></div><h1><strong>Machine learning / AI</strong></h1><p><strong>B</strong>ien, passons au plat et le machine learning, la semaine a &#233;t&#233; charg&#233;e (et la prochaine risque bien de l&#8217;&#234;tre tout autant avec des sorties du c&#244;t&#233; d&#8217;Anthropic / OpenAI) avec notamment une vid&#233;o de Peter McCormack et l&#8217;ancien &#8220;Mr AI&#8221; du Pentagone (pass&#233; dans le priv&#233; depuis), l&#8217;&#233;change est soutenu et les points &#233;voqu&#233;s sont plut&#244;t coh&#233;rents, les USA semblent comprendre que l&#8217;IA va changer bien des choses et que l&#8217;impact risque d&#8217;&#234;tre gigantesque mais doit-on s&#8217;inqui&#233;ter pour autant de tout mod&#232;le ? Dur &#224; dire, je pense que les USA se focalisent sur l&#8217;aspect &#8220;guerrier&#8221; des mod&#232;les et de l&#8217;IA, le sujet de la Chine revient souvent durant l&#8217;&#233;change (et persiste dans les discours de Mr Trump) et ce d&#233;tail m&#8217;oblige &#224; penser que les d&#233;cisions r&#233;centes de certaines startups (coucou Anthropic) quant &#224; leur entr&#233;e en bourse n&#8217;est pas totalement innocente (la FED change de discours depuis le changement de son directeur, Trump patine en Iran, la Chine livre des mod&#232;les &#224; bas co&#251;t sans rien demander en retour, cherchez le gagnant et vous trouverez celui qui l&#8217;envie).</p><p>Autre sujet avec la d&#233;couverte d&#8217;un article (via TLDR) sur l&#8217;activation &#8220;dans le futur&#8221; des mod&#232;les via un prompt syst&#232;me qui, bien exploit&#233;, pourrait inviter le mod&#232;le &#224; cr&#233;er des fichiers (voir ex&#233;cuter du code via ces m&#234;me fichiers), restons donc prudents quand il s&#8217;agit d&#8217;ex&#233;cuter des mod&#232;les hors sandbox (coucou Docker).</p><ul><li><p><a href="https://www.youtube.com/watch?v=FYRqLdpBk1g">The current state of AI, as seen by the USA</a> - Youtube</p></li><li><p><a href="https://morgin.ai/articles/your-open-source-model-could-have-a-hidden-time-release-backdoor.html">Your open-weight model could have a hidden time-release backdoor</a></p></li><li><p><a href="https://github.com/vercel-labs/fx">Fx, a Unix-like coding agent</a> - Github</p></li><li><p><a href="https://github.com/volcengine/OpenViking">OpenViking, a context database for agents</a> - Github</p></li></ul><div><hr></div><h1><strong>PHP</strong></h1><p><strong>O</strong>n termine cette veille avec PHP et la sortie de TypePHP (qui &#233;tait pr&#233;vu pour Octobre - Novembre), un compilateur AOT &#233;crit en PHP permettant de compiler du code PHP vers du C++ puis vers du natif, n&#8217;ayant pas pu me pencher sur les benchmarks face &#224; des solutions comme Elephc, je m&#8217;abstiendrais de tout avis (malgr&#233; mes contributions sur Elephc, je continue &#224; penser que toute concurrence sur ce secteur est bonne &#224; prendre, TypePHP a probablement des id&#233;es utiles pour l&#8217;&#233;cosyst&#232;me et pour PHP de fa&#231;on g&#233;n&#233;rale).</p><p>Dernier sujet avec la d&#233;couverte de TNTSearch, un moteur de recherche full-text pouvant utiliser SQLite, MySQL ou Redis (tr&#232;s utile si vous avez d&#233;j&#224; du cache / des sessions dessus).</p><ul><li><p><a href="https://mago.carthage.software/1.47.1/en/extensions/getting-started/">Mago 1.41.7 allows to write rules in plain PHP</a></p></li><li><p><a href="https://github.com/teamtnt/tntsearch">TNTSearch, a FTS search engine</a> - Github</p></li><li><p><a href="https://github.com/swoole/typephp">TypePHP, an AOT compiler for PHP</a> - Github</p></li></ul><p>Voici donc pour la veille de cette semaine, comme d&#8217;habitude, n&#8217;h&#233;sitez pas &#224; laisser un commentaire / like / partager si le coeur vous en dit, de mon c&#244;t&#233;, je vous donne rendez-vous d&#232;s la semaine prochaine pour une nouvelle veille.</p><p>D&#8217;ici-l&#224;, passez un excellent weekend ainsi qu&#8217;une tr&#232;s bonne semaine.</p><p>A vendredi.</p>]]></content:encoded></item><item><title><![CDATA[Weekly Tech - 24th of August 2026]]></title><description><![CDATA[Simulating nature in JavaScript, backing up your databases, switching from Go to Rust, a backdoor in your LLMs, and PHP diving into C++... ready?]]></description><link>https://guikingone.substack.com/p/weekly-tech-24th-of-august-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/weekly-tech-24th-of-august-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 28 Aug 2026 14:01:50 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/213058597/80cbb4fc6237b96ce5d3f5c181f7ca53.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>G</strong>ood day to everyone, I hope you are all doing well?</p><p>The Weekly Tech arrives as usual on this cloudy Friday. On the menu? General knowledge, a bit of PHP, Machine Learning, and the return of databases&#8212;all served with a few strong opinions and the usual dose of podcasts.</p><p>Anyway, let&#8217;s get straight to it. Shall we dig in?</p><div><hr></div><h1><strong>General Knowledge</strong></h1><p><strong>W</strong>e start with general knowledge, specifically the discovery of Nature of Code, a book explaining how to replicate natural systems via Javascript. I admit I&#8217;ve never really dove into some of these topics (I use vectors and neural networks daily, but actually rebuilding one from scratch is another world entirely...) and seeing code that remains &#8220;readable&#8221; and isn&#8217;t heavily reliant on external dependencies is a pleasure.</p><p>Another topic is a very short video on building a console to run the Halo series in a compact, &#8220;custom-made&#8221; format. The visual is incredibly effective, the format fits anywhere&#8212;in short, one to watch.</p><ul><li><p><a href="https://natureofcode.com/">Nature of Code, simulating natural systems in Javascript</a></p></li><li><p><a href="/__u/beefandbitcoin.substack.com/p/pourquoi-sommes-nous-autant-soumis">Why are we enslaved to the State? Testosterone might be an answer</a> - French</p></li><li><p><a href="https://www.youtube.com/watch?v=1RJcOIpp2OQ">Building the console Microsoft never will</a> - Youtube</p></li></ul><div><hr></div><h1><strong>Databases</strong></h1><p><strong>W</strong>e continue with databases, specifically Databasus, a self-hosted tool for performing backups of your favorite databases (MySQL, MariaDB, and others are also supported). It all comes with a sober but efficient UI, the ability to schedule your backups and store them off-machine&#8212;basically, very professional and also available via Docker (no more excuses).</p><ul><li><p><a href="https://github.com/databasus/databasus">Databasus, an open-source and self-hosted tool to backup PostgreSQL</a> - Github</p></li><li><p><a href="https://github.com/jeffhajewski/latticedb">LatticeDB, a property-graph database</a> - Github</p></li></ul><div><hr></div><h1><strong>Cloud Computing / Linux</strong></h1><p><strong>A</strong> small detour over to Linux for an article (which could have ended up in general knowledge, by the way) reviewing the transition from Go to Rust and the hurdles encountered along the way. Being a defender of Rust (and confessing that I&#8217;m not fond of Go), I must say the arguments hold up and are often those that come up when talking about Rust and &#8220;switching&#8221; for CLI tools and other gadgets.</p><ul><li><p><a href="https://github.com/djc/oxish">Oxish, a Rust-base memory-safe SSH server</a> - Github</p></li><li><p><a href="https://rolandsdev.blog/posts/from-go-to-rust/">From Go to Rust</a></p></li></ul><div><hr></div><h1><strong>Machine Learning / AI</strong></h1><p><strong>N</strong>ow, let&#8217;s move to the main course: machine learning. The week has been busy (and next week likely just as much with releases from Anthropic / OpenAI), notably with a video by Peter McCormack and the former &#8220;Mr AI&#8221; of the Pentagon (now in the private sector). The exchange is spirited and the points raised are quite coherent. The USA seems to understand that AI will change many things and that the impact is likely to be gigantic, but should we worry about every model? Hard to say. I think the USA is focusing on the &#8220;warfare&#8221; aspect of models and AI; the topic of China comes up often during the exchange (and persists in Mr. Trump&#8217;s discourse), and this detail makes me think that recent decisions by certain startups (hello Anthropic) regarding their IPO are not entirely innocent (the FED has changed its discourse since the change of its director, Trump is struggling in Iran, China is delivering low-cost models without asking for anything in return&#8212;look for the winner and you&#8217;ll find the one who envies them).</p><p>Another topic is the discovery of an article (via TLDR) on the &#8220;future&#8221; activation of models via a system prompt which, if well exploited, could invite the model to create files (and even execute code via those same files). So, let&#8217;s remain cautious when it comes to executing models outside of a sandbox (hello Docker).</p><ul><li><p><a href="https://www.youtube.com/watch?v=FYRqLdpBk1g">The current state of AI, as seen by the USA</a> - Youtube</p></li><li><p><a href="https://morgin.ai/articles/your-open-source-model-could-have-a-hidden-time-release-backdoor.html">Your open-weight model could have a hidden time-release backdoor</a></p></li><li><p><a href="https://github.com/vercel-labs/fx">Fx, a Unix-like coding agent</a> - Github</p></li><li><p><a href="https://github.com/volcengine/OpenViking">OpenViking, a context database for agents</a> - Github</p></li></ul><div><hr></div><h1><strong>PHP</strong></h1><p><strong>W</strong>e finish this watch with PHP and the release of TypePHP (which was planned for October - November), an AOT compiler written in PHP allowing the compilation of PHP code to C++ and then to native. Not having been able to look into the benchmarks against solutions like Elephc, I will refrain from giving an opinion (despite my contributions to Elephc, I continue to believe that any competition in this sector is welcome; TypePHP likely has useful ideas for the ecosystem and for PHP in general).</p><p>One last topic: the discovery of TNTSearch, a full-text search engine that can use SQLite, MySQL, or Redis (very useful if you already have cache / sessions on them).</p><ul><li><p><a href="https://mago.carthage.software/1.47.1/en/extensions/getting-started/">Mago 1.41.7 allows to write rules in plain PHP</a></p></li><li><p><a href="https://github.com/teamtnt/tntsearch">TNTSearch, a FTS search engine</a> - Github</p></li><li><p><a href="https://github.com/swoole/typephp">TypePHP, an AOT compiler for PHP</a> - Github</p></li></ul><p>That&#8217;s it for this Weekly Tech. As usual, feel free to leave a comment / like / share if you feel like it. For my part, I&#8217;ll see you next week for a new update.</p><p>Until then, have an excellent weekend and a great week.</p><p>See you Friday.</p>]]></content:encoded></item><item><title><![CDATA[Weekly tech - 17th of August 2026]]></title><description><![CDATA[Your devices are watching you, even in your bedroom, Mistral is shifting its goals (and its market), and PHP as a code review assistant... ready?]]></description><link>https://guikingone.substack.com/p/weekly-tech-17th-of-august-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/weekly-tech-17th-of-august-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 21 Aug 2026 14:46:13 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/212158515/38ad66fba9c34ed1de372e4f609c2a55.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p>Hello everyone, I hope you&#8217;re all doing well?</p><p>New week, new weekly watch, and this week it&#8217;s a condensed one; the week has been busy and I&#8217;ve once again lacked the time to explore certain topics (my reading list is starting to look like an endless enumeration), they will arrive late starting next week.</p><p>Anyway, let&#8217;s keep it short and sweet, shall we dive in?</p><div><hr></div><h1><strong>General Knowledge</strong></h1><p><strong>W</strong>e start with general knowledge, notably a first video by Proton which reviews the devices we use daily and which are, sometimes (often), disguised spies. It&#8217;s always curious to see certain devices (hello fridge and vibrator, what on earth is this?!) that can very quickly be diverted from their daily use, all without notification.</p><p>Another topic with an excellent video on the free market and specifically the need to regulate it; I&#8217;ll let you watch the video without giving you my opinion, it&#8217;s an essential video (especially in these troubled times and with upcoming elections).</p><p>Last topic with a very good video explaining how Ambient Occlusion works, a technique popularized by Crysis (yes, that&#8217;s getting old) allowing for the improvement of the general rendering of images generated by your GPUs. The technique has been perfected / improved and there is no doubt that many games use the Nvidia implementation rather than in-house solutions (the remastered version of Crysis uses the latter, by the way).</p><ul><li><p><a href="https://www.youtube.com/watch?v=9_WQZNEOhys">Protecting your house, by removing extra devices</a> - Youtube</p></li><li><p><a href="https://www.youtube.com/watch?v=CyhT6OfXOes">Free market don&#8217;t need regulations, at least not from the State</a> - Youtube (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=o0jBSkFELxI">Empires are doomed to fall, as Western is</a> - Youtube</p></li><li><p><a href="https://cation-lang.org/">Cation, a new functional programming language built on top of category theory</a> - Github</p></li><li><p><a href="https://www.youtube.com/watch?v=XAIfyLpxkfk">How does Ambient Occlusion works?</a> - Youtube</p></li></ul><div><hr></div><h1><strong>Machine learning / AI</strong></h1><p><strong>W</strong>e continue with machine learning / AI, notably Mistral, which is changing its vision for its future and the products it wants to promote. To be honest, I support this pivot; Mistral has understood very well that following the AI Act and various legislations, developing a model capable of competing with Claude / GPT has become impossible (thanks to the E.U. by the way...), and rather than becoming the laughingstock of the market, Mistral prefers to shift toward inference (the &#8220;act of making available&#8221; your models) and allow the execution of Chinese models (including GLM currently). This is a very good idea knowing that the servers remain on the Eurasian continent (without depending on the CCP, which is a significant plus).</p><p>Another topic with a debate on whether your models &#8220;escape the box&#8221;; I remain skeptical about this. I think models are meant &#8220;to step outside the frame,&#8221; if only to allow them to perform tasks they cannot perform &#8220;within the frame&#8221; (using tools, calculations, etc.). I maintain that &#8220;wanting to frame too much&#8221; gives the illusion of control and ends up costing very / too much when one realizes a model has escaped (thinking again of the Huggingface attack).</p><ul><li><p><a href="https://venturebeat.com/infrastructure/mistral-ai-wants-to-build-1-gigawatt-of-european-compute-by-2030-and-lock-in-customers-now">Mistral is focusing on inference</a> (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=IoM5zUI8oFc">Building a neural network to play Super Mario</a> - Youtube</p></li><li><p><a href="https://github.com/manaflow-ai/cmux">Cmux, a Ghostty-based macOS terminal for AI agents</a> - Github</p></li><li><p><a href="https://www.youtube.com/watch?v=6XrkGK9mqsE">Are Agents escaping the box? Maybe</a> - Youtube</p></li><li><p><a href="https://github.com/NanoNets/Graft">Graft, improve your agents with a graph to reduce cold start and improve quality</a> - Github</p></li><li><p><a href="https://github.com/jarrodwatts/claude-hud">A Claude Code plugin to display usage, contexte and more</a> - Github</p></li></ul><div><hr></div><h1><strong>PHP</strong></h1><p><strong>W</strong>e wrap up this week with PHP, notably an article by our national K&#233;vin Th&#233;rage on building a code review assistant via PHP, Ollama, and Qdrant. The article covers the HTTP layer, the use of the Symfony cache, the creation of commands; in short, I&#8217;ll let you read it all and try it out&#8212;you never know, it might help your PRs.</p><ul><li><p><a href="https://ktherage.github.io/fr/blog/building-a-rag-powered-code-review-assistant/">Building a code review assistant using PHP, Ollama and Qdrant</a> (French)</p></li><li><p><a href="https://symfony.com/blog/announcing-symfony-language-tools-the-official-symfony-lsp-server">A few days after Laravel, Symfony is unveiling its own LSP</a></p></li></ul><p>That&#8217;s it for this week&#8217;s watch. As usual, feel free to leave a comment / like / share if you feel like it. For my part, I&#8217;ll see you next week for a new update.</p><p>Until then, have a great weekend and a wonderful week.</p><p>See you Friday.</p>]]></content:encoded></item><item><title><![CDATA[Veille de la semaine du 17 Août 2026]]></title><description><![CDATA[Vos appareils vous surveillent jusqu&#8217;&#224; dans votre chambre, Mistral change d&#8217;objectif (et donc de march&#233;) et PHP comme assistant de review, pr&#234;t(e)s ?]]></description><link>https://guikingone.substack.com/p/veille-de-la-semaine-du-17-aout-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/veille-de-la-semaine-du-17-aout-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 21 Aug 2026 14:46:12 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/212157864/3b482da9e7c1e4de472824d597618805.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>B</strong>ien le bonjour &#224; tous et &#224; toutes, j&#8217;esp&#232;re que tout le monde va bien ?</p><p>Nouvelle semaine, nouvelle Veille de la semaine et cette semaine, elle se veut condens&#233;e, la semaine a &#233;t&#233; charg&#233;e et j&#8217;ai encore manqu&#233; de temps pour explorer certains sujets (ma liste de lecture commence &#224; rassembler &#224; une &#233;num&#233;ration sans fin), ils arriveront avec du retard d&#232;s la semaine prochaine.</p><p>Bref, parlons peu, parlons bien, on se lance ?</p><div><hr></div><h1><strong>Culture g&#233;n&#233;rale</strong></h1><p><strong>O</strong>n commence donc par la culture g&#233;n&#233;rale et notamment une premi&#232;re vid&#233;o par Proton qui revient sur les appareils que l&#8217;on utilise au quotidien et qui sont, parfois (souvent), des mouchards d&#233;guis&#233;s, toujours curieux de voir certains appareils (coucou le frigo et le vibromasseur, qu&#8217;est-ce donc que ceci ?!) qui peuvent tr&#232;s vite &#234;tre d&#233;tourn&#233;s de leur usage quotidien, le tout sans notification.</p><p>Autre sujet avec une excellente vid&#233;o sur le march&#233; libre et notamment le fait de devoir r&#233;guler ce dernier, je vous laisse regarder la vid&#233;o sans vous donner mon avis, c&#8217;est une vid&#233;o indispensable (surtout en ces temps troubl&#233;s et avec des &#233;lections &#224; venir).</p><p>Dernier sujet avec une tr&#232;s bonne vid&#233;o expliquant comme fonctionne l&#8217;Oclusion Ambiante, technique popularis&#233;e par Crysis (oui, &#231;a commence &#224; dater) permettant d&#8217;am&#233;liorer le rendu g&#233;n&#233;ral des images g&#233;n&#233;r&#233;es par vos GPU&#8217;s, la technique a &#233;t&#233; perfectionn&#233;e / am&#233;lior&#233;e et nulle doute que beaucoup de jeux utilisent l&#8217;impl&#233;mentation Nvidia plut&#244;t que des solutions maison (la version remastered de Crysis utilise cette derni&#232;re d&#8217;ailleurs).</p><ul><li><p><a href="https://www.youtube.com/watch?v=9_WQZNEOhys">Protecting your house, by removing extra devices</a> - Youtube</p></li><li><p><a href="https://www.youtube.com/watch?v=CyhT6OfXOes">Free market don&#8217;t need regulations, at least not from the State</a> - Youtube (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=o0jBSkFELxI">Empires are doomed to fall, as Western is</a> - Youtube</p></li><li><p><a href="https://cation-lang.org/">Cation, a new functional programming language built on top of category theory</a> - Github</p></li><li><p><a href="https://www.youtube.com/watch?v=XAIfyLpxkfk">How does Ambient Occlusion works?</a> - Youtube</p></li></ul><div><hr></div><h1><strong>Machine learning / AI</strong></h1><p><strong>O</strong>n continue avec le machine learning / AI et notamment Mistral qui change de vision sur son avenir et les produits qu&#8217;elle souhaite mettre en avant, pour &#234;tre franc, je soutiens cette bascule, Mistral a tr&#232;s bien compris que suite &#224; l&#8217;AI Act et aux diff&#233;rentes l&#233;gislations, d&#233;velopper un mod&#232;le pouvant concurrencer Claude / GPT est devenu impossible (merci l&#8217;U.E au passage ...), &#224; devenir la ris&#233;e du march&#233;, Mistral pr&#233;f&#232;re basculer sur l&#8217;inf&#233;rence (le &#8220;fait de rendre disponible&#8221; vos mod&#232;les) et permettre d&#8217;ex&#233;cuter des mod&#232;les chinois (dont GLM &#224; date), une tr&#232;s bonne id&#233;e sachant que les serveurs restent sur le continent eurasiatique (sans d&#233;pendre du PCC, c&#8217;est un plus non n&#233;gligeable).</p><p>Autre sujet avec un d&#233;bat sur le fait de savoir si vos mod&#232;les &#8220;sortent de la boite&#8221;, je reste sceptique &#224; ce sujet, je pense que les mod&#232;les ont vocation &#8220;&#224; sortir du cadre&#8221;, ne serait-ce que pour leur permettre d&#8217;effectuer des t&#226;ches qu&#8217;ils ne peuvent pas effectuer &#8220;dans le cadre&#8221; (utiliser des outils, des calculs, etc), je persiste &#224; penser que &#8220;vouloir trop cadrer&#8221; donne l&#8217;illusion du contr&#244;le et finit par co&#251;ter tr&#232;s / trop cher quand on s&#8217;aper&#231;oit qu&#8217;un mod&#232;le s&#8217;&#233;chappe (on pense encore &#224; l&#8217;attaque ur Hugginggface).</p><ul><li><p><a href="https://venturebeat.com/infrastructure/mistral-ai-wants-to-build-1-gigawatt-of-european-compute-by-2030-and-lock-in-customers-now">Mistral is focusing on inference</a> (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=IoM5zUI8oFc">Building a neural network to play Super Mario</a> - Youtube</p></li><li><p><a href="https://github.com/manaflow-ai/cmux">Cmux, a Ghostty-based macOS terminal for AI agents</a> - Github</p></li><li><p><a href="https://www.youtube.com/watch?v=6XrkGK9mqsE">Are Agents escaping the box? Maybe</a> - Youtube</p></li><li><p><a href="https://github.com/NanoNets/Graft">Graft, improve your agents with a graph to reduce cold start and improve quality</a> - Github</p></li><li><p><a href="https://github.com/jarrodwatts/claude-hud">A Claude Code plugin to display usage, contexte and more</a> - Github</p></li></ul><div><hr></div><h1><strong>PHP</strong></h1><p><strong>O</strong>n termine cette semaine avec PHP et notamment un article de notre K&#233;vin Th&#233;rage national sur la construction d&#8217;un assistant de code review via PHP, Ollama et Qdrant, l&#8217;article passe par la couche HTTP, l&#8217;utilisation du cache Symfony, la cr&#233;ation de commandes, bref, je vous laisse lire le tout et l&#8217;essayer, on ne sait jamais, cela pourrait aider vos PRs.</p><ul><li><p><a href="https://ktherage.github.io/fr/blog/building-a-rag-powered-code-review-assistant/">Building a code review assistant using PHP, Ollama and Qdrant</a> (French)</p></li><li><p><a href="https://symfony.com/blog/announcing-symfony-language-tools-the-official-symfony-lsp-server">A few days after Laravel, Symfony is unveiling its own LSP</a></p></li></ul><p>Voici donc pour la veille de cette semaine, comme d&#8217;habitude, n&#8217;h&#233;sitez pas &#224; laisser un commentaire / like / partager si le coeur vous en dit, de mon c&#244;t&#233;, je vous donne rendez-vous d&#232;s la semaine prochaine pour une nouvelle veille.</p><p>D&#8217;ici-l&#224;, passez un excellent weekend ainsi qu&#8217;une tr&#232;s bonne semaine.</p><p>A vendredi.</p>]]></content:encoded></item><item><title><![CDATA[Weekly Tech - 10th of August 2026]]></title><description><![CDATA[AI is disrupting our projects&#8212;will it disrupt our lives? Can Symfony export and import XLSX files? A touch of open-source S3 and AI as a pentester... ready?]]></description><link>https://guikingone.substack.com/p/weekly-tech-10th-of-august-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/weekly-tech-10th-of-august-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 14 Aug 2026 14:31:10 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/211185126/1510066f7268fb7f32f5d70e355348f5.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>H</strong>ello everyone, I hope you are all doing well?</p><p>New week, new Weekly Tech, and this week, there is a stronger focus on machine learning / AI as well as general knowledge. Some topics took me longer than expected during testing, and I prefer to push their publication back by a week (or two) to provide a more reasoned opinion&#8212;you know how it goes.</p><p>Anyway, let&#8217;s get straight to it. Shall we?</p><div><hr></div><h1><strong>General Knowledge</strong></h1><p><strong>W</strong>ell, let&#8217;s start with general knowledge. You&#8217;re starting to see a pattern: I use this block to slide in topics that go beyond tech, often podcasts from which I extract lessons and/or subjects that follow me over several weeks. This week, I listened to an interview/exchange on AI and its impact on the job market as well as other sectors. Some positions may seem &#8220;extreme,&#8221; I readily admit, but I think some stances are valid. AI will probably &#8220;change&#8221; our vision of the world and even reshape it when the &#8220;general public&#8221; gets involved (for now, outside of &#8220;tech&#8221; circles, the general public mostly plays with Gemini or ChatGPT, so the majority have only a truncated view&#8212;when they have one at all). Can we truly estimate the impact? Hard to say; the invention of the assembly line in Henry Ford&#8217;s time was probably not fully measured when it first appeared.</p><ul><li><p><a href="https://www.youtube.com/watch?v=uLLO1LbscCE">The digital Euro is a trap</a> - Youtube (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=PJyCHMpdkIc">Is AI the next big crisis? Maybe</a> - Youtube (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=j4x5HjbFVTc">Is Switzerland really neutral?</a> - Youtube (French)</p></li></ul><div><hr></div><h1><strong>Databases</strong></h1><p><strong>A</strong> quick detour into databases to discover FalkorDB (I could have slipped it into ML / AI, but hey, let&#8217;s be crazy and leave it here; it doesn&#8217;t cost anything), a graph-oriented database. The marketing mainly steers it toward AI (because that&#8217;s the trend these days), but I still struggle to see the immediate benefit of this type of database. It often requires labeling relationships yourself (in the era of LLMs, this can be automated but with certain constraints), and I&#8217;ve often encountered performance issues with large volumes of data or slightly complex code.</p><p>In short, Falkor proposes itself for managing the memory of your agents. I haven&#8217;t had the chance to test it in that context yet, so we&#8217;ll see how it goes.</p><ul><li><p><a href="https://github.com/FalkorDB/FalkorDB">FalkorDB, a graph database for LLMs</a> - Github</p></li></ul><div><hr></div><h1><strong>Cloud Computing / Linux</strong></h1><p><strong>R</strong>eturn to basics with &#8220;cloud&#8221; and the discovery of Celld, an S3 alternative based solely on SQLite (as I&#8217;ve told you before, SQLite is never far away). The library is open-source, so you have every opportunity to deploy it wherever suits you (even within your agents...) without depending on AWS.</p><ul><li><p><a href="https://github.com/denoland/celld">Celld, a self-hosted distributed durable object engine built on top of SQLite</a> - Github</p></li><li><p><a href="https://github.com/vicinaehq/vicinae">Vicinae, a focused launcher for your desktop</a> - Github</p></li></ul><div><hr></div><h1><strong>Machine Learning / AI</strong></h1><p><strong>B</strong>ack to the classics with machine learning, specifically Sayit, a library for macOS (I know, sorry) that allows (as Handy also does) the use of TTS models to give &#8220;voice&#8221; to your agents and simplify your exchanges with them. I haven&#8217;t had the chance to test it (Handy remains the installed solution that fits my daily needs), but the library is built around MLX, an approach designed for macOS that allows you to load your models while leveraging the ecosystem, promising much better performance than other market alternatives.</p><p>Another topic: a podcast in which the creator of Google X appeared to debate whether the famous singularity has been reached (or will be soon). Many subjects have been impacted by AI to date, and even more will be tomorrow; it remains to be seen if we are capable of adapting.</p><ul><li><p><a href="https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content">Claude will include watermarks in generated content</a></p></li><li><p><a href="https://www.youtube.com/watch?v=7Nu39IzAAq4">Are we close to the singularity?</a> - Youtube</p></li><li><p><a href="https://github.com/callebtc/sayit">Sayit, a local / private TTS for Apple Silicon Macs</a> - Github</p></li><li><p><a href="https://blog.casa.io/how-built-reliable-ai-pentest/">Building a reliable pentesting agent</a></p></li><li><p><a href="https://www.intheweights.com/">Are you in the weights?</a></p></li></ul><div><hr></div><h1><strong>PHP</strong></h1><p><strong>A</strong> very brief detour to PHP to discover a Symfony bundle that allows exporting/importing your entities to Excel / CSV formats. The support seems comprehensive (especially for enumerations and Doctrine types) and the bundle is quite configurable. Having not had the chance to test it (busy week on my end), I&#8217;ll let you form your own opinion.</p><ul><li><p><a href="https://github.com/HugoSEIGLE/symfony-import-export-bundle">A Symfony bundle to import / export entities to Excel / CSV files</a> - Github</p></li></ul><div><hr></div><h1><strong>Tools</strong></h1><p><strong>F</strong>inal topic: an open-source tool using Bluetooth to find your smartphones. Very useful if no software solution exists or if they are being stubborn. A very cool idea to make it easier to find smartphones forgotten in the open-space (we see you).</p><ul><li><p><a href="https://github.com/ben-z/findphone">FindPhone, locate Bluetooth device by signal strength</a> - Github</p></li></ul><p>That&#8217;s it for this Weekly Tech. As usual, feel free to leave a comment / like / share if you feel like it. For my part, I&#8217;ll see you next week for a new update.</p><p>Until then, have a great weekend and a wonderful week.</p><p>See you Friday.</p>]]></content:encoded></item><item><title><![CDATA[Veille de la semaine du 10 Août 2026]]></title><description><![CDATA[L'IA bouleverse nos projets, bouleversera t-elle nos vies? Symfony peut-il exporter/importer du XLSX? Un brin de S3 open-source et l'IA comme pentester, pr&#234;t(e)s ?]]></description><link>https://guikingone.substack.com/p/veille-de-la-semaine-du-10-aout-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/veille-de-la-semaine-du-10-aout-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 14 Aug 2026 14:31:07 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/211183945/9ad6331a21bd5112f2700b072256d9ba.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>B</strong>ien le bonjour &#224; tous et &#224; toutes, j&#8217;esp&#232;re que tout le monde va bien ?</p><p>Nouvelle semaine, nouvelle Veille de la semaine et cette semaine, focus plus marqu&#233; sur le machine learning / AI ainsi que sur la culture g&#233;n&#233;rale, des sujets m&#8217;ont pris plus de temps que pr&#233;vu durant les tests et je pr&#233;f&#232;re d&#233;caler d&#8217;une semaine (voir deux) leur publication afin d&#8217;avoir un avis plus argument&#233;, vous commencez &#224; connaitre la chanson.</p><p>Bref, parlons peu, parlons bien, on se lance ?</p><div><hr></div><h1><strong>Culture g&#233;n&#233;rale</strong></h1><p><strong>B</strong>ien, commen&#231;ons par la culture g&#233;n&#233;rale, vous commencez &#224; voir se dessiner un pattern, j&#8217;utilise ce bloc pour y glisser les sujets qui d&#233;passent la tech, souvent des podcasts sur lesquels j&#8217;extrait des enseignements et/ou des sujets qui me suivent sur plusieurs semaines, cette semaine, j&#8217;ai pu &#233;couter une interview / &#233;change sur l&#8217;IA et son impact sur le march&#233; du travail ainsi que sur les autres secteurs, certaines positions peuvent sembler &#8220;extr&#234;mes&#8221;, je le conc&#232;de ais&#233;ment mais je pense que certaines prises de position se tiennent, l&#8217;IA va probablement &#8220;changer&#8221; notre vision du monde et m&#234;me remodeler ce dernier quand le &#8220;grand public&#8221; s&#8217;y mettra (pour le moment et sorti du cadre des &#8220;techs&#8221;, le grand public joue majoritairement avec Gemini ou ChatGPT, ils n&#8217;ont donc qu&#8217;une vision tronqu&#233;e pour la majorit&#233; - quand ils en ont une -), peut-on r&#233;ellement estimer l&#8217;impact ? Difficile &#224; dire, l&#8217;invention de la cha&#238;ne de fabrication / montage &#224; l&#8217;&#233;poque d&#8217;Henry Ford n&#8217;&#233;tait probablement pas mesur&#233;e quand elle a vu le jour.</p><ul><li><p><a href="https://www.youtube.com/watch?v=uLLO1LbscCE">The digital Euro is a trap</a> - Youtube (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=PJyCHMpdkIc">Is AI the next big crisis? Maybe</a> - Youtube (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=j4x5HjbFVTc">Is Switzerland really neutral?</a> - Youtube (French)</p></li></ul><div><hr></div><h1><strong>Base de donn&#233;es</strong></h1><p><strong>T</strong>out petit d&#233;tour du c&#244;t&#233; de la base de donn&#233;es pour d&#233;couvrir FalkorDB (j&#8217;aurais pu le glisser dans ML / AI mais bon, soyons fous et laissons-le ici, &#231;a ne co&#251;te rien), une base de donn&#233;es orient&#233;e graph, le marketing l&#8217;oriente principalement vers l&#8217;IA (parce que c&#8217;est devenu une mode de nos jours) mais je persiste &#224; avoir du mal &#224; voir l&#8217;int&#233;r&#234;t imm&#233;diat de ce type de base de donn&#233;es, cela n&#233;cessite souvent d&#8217;&#233;tiqueter soi-m&#234;me les relations (&#224; l&#8217;&#232;re des LLMs, on peut automatiser mais avec certaines contraintes) et j&#8217;ai souvent eu des probl&#232;mes de performances sur des larges volumes de donn&#233;es / du code un poil complexe.</p><p>Bref, Falkor se propose pour la gestion de la m&#233;moire de vos agents, je n&#8217;ai pas eu l&#8217;occasion de le tester dans ce cadre, &#224; voir ce que cela donne.</p><ul><li><p><a href="https://github.com/FalkorDB/FalkorDB">FalkorDB, a graph database for LLMs</a> - Github</p></li></ul><div><hr></div><h1><strong>Cloud computing / Linux</strong></h1><p><strong>B</strong>ien, retour aux fondamentaux avec le &#8220;cloud&#8221; et la d&#233;couverte de Celld, une alternative &#224; S3 se basant que SQLite (je vous l&#8217;ai d&#233;j&#224; dit, SQLite n&#8217;est jamais loin), la librairie est open-source, vous avez donc tout loisir de d&#233;ployer le tout o&#249; cela vous arrange (et ce, m&#234;me dans vos agents ... ) sans d&#233;pendre d&#8217;AWS.</p><ul><li><p><a href="https://github.com/denoland/celld">Celld, a self-hosted distributed durable object engine built on top of SQLite</a> - Github</p></li><li><p><a href="https://github.com/vicinaehq/vicinae">Vicinae, a focused launcher for your desktop</a> - Github</p></li></ul><div><hr></div><h1><strong>Machine learning / AI</strong></h1><p><strong>R</strong>etour aux classiques avec le machine learning et notamment Sayit, une librairie pour macOS (je sais, d&#233;sol&#233;) permettant (comme Handy le permet d&#8217;ailleurs) d&#8217;utiliser des mod&#232;les TTS afin de donner &#8220;la parole&#8221; &#224; vos agents et simplifier vos &#233;changez avec ces derniers, je n&#8217;ai pas eu l&#8217;occasion de le tester (Handy restant la solution install&#233;e et convenant &#224; mes besoins quotidiens) mais la librairie est construite autour de MLX, une approche pens&#233;e pour macOS permettant de charger vos mod&#232;les en profitant de l&#8217;&#233;cosyst&#232;me, cela promet donc des performances bien meilleurs que les alternatives du march&#233;.</p><p>Autre sujet avec un podcast dans lequel le cr&#233;ateur de Google X intervenait afin de d&#233;battre du fait que la fameuse singularit&#233; aurait &#233;t&#233; atteinte (ou le serait prochainement), bien des sujets ont &#233;t&#233; impact&#233;s par l&#8217;IA &#224; date et encore plus le seront demain, reste &#224; voir si nous sommes capables de nous adapter.</p><ul><li><p><a href="https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content">Claude will include watermarks in generated content</a></p></li><li><p><a href="https://www.youtube.com/watch?v=7Nu39IzAAq4">Are we close to the singularity?</a> - Youtube</p></li><li><p><a href="https://github.com/callebtc/sayit">Sayit, a local / private TTS for Apple Silicon Macs</a> - Github</p></li><li><p><a href="https://blog.casa.io/how-built-reliable-ai-pentest/">Building a reliable pentesting agent</a></p></li><li><p><a href="https://www.intheweights.com/">Are you in the weights?</a></p></li></ul><div><hr></div><h1><strong>PHP</strong></h1><p><strong>T</strong>r&#232;s l&#233;ger d&#233;tour du c&#244;t&#233; de PHP pour d&#233;couvrir un bundle Symfony permettant d&#8217;exporter / importer vos entit&#233;s vers les formats Excel / CSV, le support semble complet (notamment les &#233;num&#233;rations et types Doctrine) et le bundle plut&#244;t configurable, n&#8217;ayant pas eu l&#8217;occasion de le tester (semaine charg&#233;e de mon c&#244;t&#233;), je vous laisse vous faire un avis.</p><ul><li><p><a href="https://github.com/HugoSEIGLE/symfony-import-export-bundle">A Symfony bundle to import / export entities to Excel / CSV files</a> - Github</p></li></ul><div><hr></div><h1><strong>Tools</strong></h1><p><strong>D</strong>ernier sujet avec un outil open-source utilisant le Bluetooth afin de retrouver vos smartphones, tr&#232;s utile si aucune solution logicielle n&#8217;existe ou si ces derni&#232;res vous tiennent t&#234;te, tr&#232;s chouette id&#233;e afin de permettre plus facilement de retrouver vos smartphones oubli&#233;s dans l&#8217;open-space (on vous voit).</p><ul><li><p><a href="https://github.com/ben-z/findphone">FindPhone, locate Bluetooth device by signal strength</a> - Github</p></li></ul><p>Voici donc pour la veille de cette semaine, comme d&#8217;habitude, n&#8217;h&#233;sitez pas &#224; laisser un commentaire / like / partager si le coeur vous en dit, de mon c&#244;t&#233;, je vous donne rendez-vous d&#232;s la semaine prochaine pour une nouvelle veille.</p><p>D&#8217;ici-l&#224;, passez un excellent weekend ainsi qu&#8217;une tr&#232;s bonne semaine.</p><p>A vendredi.</p>]]></content:encoded></item><item><title><![CDATA[Veille de la semaine du 3 Août 2026]]></title><description><![CDATA[Vos base de donn&#233;es m&#233;ritent aussi du test de mont&#233;e en charge, l'IA serait-elle (non) une bulle ? Conserver vos notifications malgr&#233; le cache HTTP et bien d'autres joyeuset&#233;es, pr&#234;t(e)s ?]]></description><link>https://guikingone.substack.com/p/veille-de-la-semaine-du-3-aout-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/veille-de-la-semaine-du-3-aout-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 07 Aug 2026 14:30:51 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/210226781/298f5fd2bcc6316cb5b2e85e3c4a0314.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>B</strong>ien le bonjour &#224; tous et &#224; toutes, j&#8217;esp&#232;re que tout le monde va bien ?</p><p>Nouvelle semaine, nouvelle Veille de la semaine, &#224; l&#8217;image de ces derni&#232;res semaines, elles se veut plus l&#233;g&#232;re, la p&#233;riode est calme, je rattrape p&#233;niblement mon retard sur certaines lectures et certains sujets tombent parfois sans pr&#233;venir, bref, vous commencez &#224; connaitre la chanson.</p><p>On se lance ?</p><div><hr></div><h1>Culture g&#233;n&#233;rale</h1><p><strong>O</strong>n commence par la culture g&#233;n&#233;rale et notamment quelques vid&#233;os, je m&#8217;arr&#234;te volontairement sur une vid&#233;o qui a fait du bruit r&#233;cemment, vous l&#8217;avez probablement vue, je parle de l&#8217;interview d&#8217;Elon Musk chez The Economist, comme &#224; l&#8217;accoutum&#233;e, je ne parlerais pas de la personne qu&#8217;est Elon Musk car je pr&#233;f&#232;re me concentrer sur le propos (et que donner mon avis ne changera rien aux propos tenus durant l&#8217;&#233;change), certains sujets abord&#233;s concerne tout le monde, d&#8217;autres peuvent ouvrir au d&#233;bat sur la fa&#231;on dont ils sont pr&#233;sent&#233;s, je vous laisse vous forger un avis.</p><p>Autre sujet avec la faille du c&#244;t&#233; de Coldcard qui a fait beaucoup de bruit (&#224; raison), pour rappel, ce hardware wallet (qui se disait le plus s&#233;curis&#233; au monde - on note l&#8217;exc&#232;s de confiance -) a failli lors de la g&#233;n&#233;ration de la clef priv&#233;e (et ce, depuis des ann&#233;es) et a donc mis en danger des milliers de personnes. Pour vulgariser l&#8217;id&#233;e, Coldcard s&#8217;engageait &#224; g&#233;n&#233;rer une part d&#8217;al&#233;atoire afin de faire &#8220;rouler les d&#233;s&#8221; lors de la g&#233;n&#233;ration de votre clef priv&#233;e, en th&#233;orie, Coldcard &#233;tait cens&#233; utiliser 256 bits comme entropie afin d&#8217;obtenir une clef qui n&#233;cessitait des milliards de milliards de millards (et encore plus) de tentatives pour &#234;tre trouv&#233;es, le hic &#233;tant que le firmware d&#233;ploy&#233; en 2021 n&#8217;utilisait que 40 bits comme entropie (et ce, &#224; cause d&#8217;un bug qui a pass&#233; les &#8220;tests&#8221; - dont on commence &#224; voir qu&#8217;ils n&#8217;existaient pas vraiment -), ce qui r&#233;duisait drastiquement la complexit&#233; des calculs requis pour &#8220;retrouver&#8221; la clef priv&#233;e. Cela a donc conduit &#224; une s&#233;rie d&#8217;attaque sur des adresses d&#233;tenant des milliers de bitcoins, pour sch&#233;matiser, imaginer que quelqu&#8217;un arrive &#224; voler le contenu d&#8217;un coffre sans l&#8217;ouvrir ni sans &#234;tre physiquement devant ce dernier, des patchs ont &#233;t&#233; d&#233;ploy&#233;s mais le mal est fait et des centaines de personnes se retrouvent d&#233;poss&#233;d&#233;es de leur bitcoins.</p><p>Rappel utile, cette faille n&#8217;est pas li&#233;e &#224; Bitcoin, le protocol n&#8217;a pas chang&#233; et n&#8217;a pas &#233;t&#233; attaqu&#233;, on parle d&#8217;une attaque sur un &#8220;fournisseur de solution&#8221;, cela remet en question cet &#233;cosyst&#232;me mais pas Bitcoin (distinction qui a son importance, certains r&#233;gulateurs &#233;tant tr&#232;s prompt &#224; se concentrer sur Bitcoin &#8220;parce que&#8221; plut&#244;t que sur le vrai probl&#232;me).</p><ul><li><p><a href="https://www.youtube.com/watch?v=XuoqKYxDHVc">An interview with Elon Musk</a> - Youtube</p></li><li><p><a href="https://wizardsardine.com/blog/coldcard-rng-vulnerability/">The Coldcard flaw is bad, really bad and you should be worried</a></p></li><li><p><a href="https://itsallaboutthebit.com/memory-safety-absolutists/">Is memory safety really important?</a></p></li><li><p><a href="https://www.youtube.com/watch?v=VeAfJfxoFac">In case you have doubts, the war against Iran wasn&#8217;t about Iran and/or nuclear</a> - Youtube</p></li></ul><div><hr></div><h1>Base de donn&#233;es</h1><p><strong>O</strong>n continue avec la base de donn&#233;es et notamment Barrage, un outil de load testing, qui permet de tester les applications web mais aussi la base de donn&#233;es. Barrage vous permet donc de v&#233;rifier les performances de votre application tout en soumettant votre base de donn&#233;es (SQL et/ou Redis-like au passage) &#224; des stress tests, un indispensable si vous vous retrouvez r&#233;guli&#232;rement confront&#233;s &#224; des probl&#232;mes de performances o&#249; l&#8217;application semble lente mais que vos outils type Blackfire ou consorts ne remontent rien, cela peut permettre de d&#233;tecter des indexes non utilis&#233;s, des soucis de sch&#233;mas voir des probl&#232;mes de cache, le tout en profitant d&#8217;un dashboard afin de suivre l&#8217;avanc&#233;e des tests et l&#8217;impact sur la stack sous charge.</p><p>Un indispensable &#224; glisser dans vos boites &#224; outils pour vos futurs sessions de tests.</p><ul><li><p><a href="https://github.com/singhpratech/ferrovec">Ferrovec, a WASM-powered vector store</a> - Github</p></li><li><p><a href="https://github.com/codetesla51/barrage">Barrage, a load testing tool for applications and databases</a> - Github</p></li></ul><div><hr></div><h1>Cloud computing / Linux</h1><p><strong>T</strong>out petit d&#233;tour du c&#244;t&#233; du cloud pour d&#233;couvrir BrewFS, un filesystem distribu&#233; &#8220;haute performance&#8221; majoritairement mis en avant pour les agents AI et le stockage type S3, vous avez probablement d&#233;j&#224; &#233;t&#233; confront&#233; &#224; des probl&#232;mes de performances lorsque plusieurs agents veulent acc&#233;der au m&#234;me projet / fichier, BrewFS propose une solution &#224; ce type de probl&#232;me (et bien d&#8217;autres) et permet &#224; vos agents de profiter d&#8217;un FS compatible avec d&#8217;autres solutions type Redis, SQLite et consorts.</p><ul><li><p><a href="https://github.com/brewfs/brewfs">BrewFS, a high-performance distributed filesystem for AI, object storage and more</a> - Github</p></li></ul><div><hr></div><h1>Machine learning / AI</h1><p><strong>O</strong>n revient aux classiques avec le machine learning / AI et notamment Swarm Forge, une &#8220;forge&#8221; pour vos agents afin de piloter et structurer leur travaux, le tout, construit par Oncle Bob (je me garderais bien de donner un avis sur le personnage, je regarde son travail et j&#8217;appr&#233;cie ce qu&#8217;il a mis en avant dans l&#8217;&#233;cosyst&#232;me, le reste ne me regarde pas). J&#8217;avoue appr&#233;cier ce genre de projet notamment car les outils de harness type Codex / Claude Code ne proposent pas ce type de gestion granulaire de vos agents / sous-agents, ils lancent ces derniers mais ne donne que tr&#232;s peu (voir aucun) contr&#244;le sur lesdits agents, on se retrouve donc tr&#232;s souvent &#224; &#8220;r&#233;primander&#8221; les agents pour qu&#8217;ils restent dans un cadre ou qu&#8217;ils en sortent lorsque c&#8217;est n&#233;cessaire et on finit par perdre &#233;norm&#233;ment de temps car un agent &#8220;ne rentre pas dans la bonne case&#8221;, l&#8217;approche n&#8217;est pas r&#233;volutionnaire mais permet de mieux cadrer vos agents et &#233;viter des d&#233;bordement que les harness ne d&#233;tectent pas (ou d&#233;cident de laisser couler).</p><p>Autre point avec une vid&#233;o YouTube du c&#244;t&#233; de WBD avec Luke Gromen qui s&#8217;interroge sur la viabilit&#233; du business model de l&#8217;IA et des soci&#233;t&#233;s qui font avancer cet &#233;cosyst&#232;me, l&#8217;IA n&#8217;est probablement pas une bulle (du moins, pas au sens financier du terme) mais les soci&#233;t&#233;s au sein de cet &#233;cosyst&#232;me posent probl&#232;me, leur financement est souvent en circuit ferm&#233; (bien que tr&#232;s d&#233;pendant de la FED en bout de course) et elles s&#8217;auto-entretiennent dans un cercle d&#8217;achat / vente / actions pour financer la course &#224; l&#8217;AGI, r&#233;sultat, l&#8217;&#233;cosyst&#232;me tourne sur lui-m&#234;me et il y a fort &#224; parier que sans le dollar, rien ne garanti qu&#8217;il survivrait.</p><ul><li><p><a href="https://github.com/unclebob/swarm-forge">SwarmForge, a tool to coordinate agents</a> - Github</p></li><li><p><a href="https://www.youtube.com/watch?v=w-Vopzt_m5g">AI is a different kind of bubble</a> - Youtube</p></li><li><p><a href="https://www.bitpanda.com/en/ai-investing-assistant">Bitpanda is opening its own MCP server</a></p></li></ul><div><hr></div><h1>PHP</h1><p><strong>O</strong>n termine avec PHP et notamment un bundle qui permet de g&#233;n&#233;rer des notifications de type toast sur vos applications, le tout en conservant le syst&#232;me de cache HTTP (si activ&#233;), un probl&#232;me que l&#8217;on rencontre tr&#232;s souvent dans nos applications (apr&#232;s tout, une notification ne co&#251;te rien, autant en profiter) mais qui se heurte &#224; ce fameux mur du cache HTTP (autre vaste sujet), le bundle utilise Turbo Stream / Stimulus afin de contourner le bloquage et permet de conserver vos notifications en toute situation.</p><p>Tout dernier sujet avec QueueSQL, une librairie pour Laravel permettant de stocker dans des queues vos requ&#234;tes en &#233;criture, pour les amoureux de Symfony, cela revient &#224; utiliser un bus d&#233;di&#233; dans Messenger avec un stockage via RabbitMQ / Redis / Doctrine avec une architecture propre de CQRS (j&#8217;aurais beaucoup &#224; dire au sujet de ce dernier), QueueSQL permettant &#233;galement l&#8217;utilisation d&#8217;un &#8220;parall&#233;lisme&#8221; (en pratique, PSL et Parallel le permettent aussi au besoin).</p><ul><li><p><a href="https://symfony.com/blog/introducing-symfony-reprise-the-symfony-integration-layer-for-modern-bundlers">Symfony is unveiling Reprise, a modern layer for bundlers</a></p></li><li><p><a href="https://github.com/marilenaRM/turbo-toast-bundle">A bundle to trigger toast notification while keeping HTTP cache</a> - Github</p></li><li><p><a href="https://github.com/kamranata/queue-sql">QueueSQL, a library to queue any Laravel write through parallel batched jobs</a> - Github</p></li></ul><p>Voici donc pour la veille de cette semaine, comme d&#8217;habitude, n&#8217;h&#233;sitez pas &#224; laisser un commentaire / like / partager si le coeur vous en dit, de mon c&#244;t&#233;, je vous donne rendez-vous d&#232;s la semaine prochaine pour une nouvelle veille.</p><p>D&#8217;ici-l&#224;, passez un excellent weekend ainsi qu&#8217;une tr&#232;s bonne semaine.</p><p>A vendredi.</p>]]></content:encoded></item><item><title><![CDATA[Weekly Tech - 3 August 2026]]></title><description><![CDATA[Your databases deserve load testing too, is AI a bubble (or not)? Keeping your notifications alive despite HTTP caching, and plenty of other goodies... ready?]]></description><link>https://guikingone.substack.com/p/weekly-tech-3-august-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/weekly-tech-3-august-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 07 Aug 2026 14:30:35 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/210227611/305d9b43f000cab5f2c5d52faa2f0913.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>G</strong>ood day to everyone, I hope you are all doing well?</p><p>New week, new weekly watch. Like the last few weeks, this one is lighter; things are calm, I&#8217;m slowly catching up on some reading, and some topics occasionally pop up without warning&#8212;in short, you know the drill.</p><p>Shall we dive in?</p><div><hr></div><h1>General Knowledge</h1><p><strong>W</strong>e start with general knowledge, specifically a few videos. I&#8217;m intentionally stopping on a video that has caused quite a stir recently&#8212;you&#8217;ve probably seen it: the interview with Elon Musk at The Economist. As usual, I won&#8217;t talk about Elon Musk as a person because I prefer to focus on the content (and because giving my opinion won&#8217;t change the statements made during the exchange). Some of the topics discussed concern everyone, while others may open a debate on how they are presented; I&#8217;ll let you form your own opinion.</p><p>Another topic is the flaw on the Coldcard side that caused a lot of noise (rightly so). As a reminder, this hardware wallet (which claimed to be the most secure in the world&#8212;note the overconfidence) failed during the private key generation (and has done so for years), thereby endangering thousands of people. To simplify the idea, Coldcard committed to generating a portion of randomness to &#8220;roll the dice&#8221; during your private key generation. In theory, Coldcard was supposed to use 256 bits of entropy to obtain a key that would require billions upon billions of attempts to be found. The catch was that the firmware deployed in 2021 used only 40 bits of entropy (due to a bug that passed the &#8220;tests&#8221;&#8212;which we are starting to see didn&#8217;t really exist), drastically reducing the complexity of the calculations required to &#8220;find&#8221; the private key. This led to a series of attacks on addresses holding thousands of bitcoins. To put it simply, imagine someone being able to steal the contents of a safe without opening it or even being physically in front of it. Patches have been deployed, but the damage is done, and hundreds of people have been stripped of their bitcoins.</p><p>A useful reminder: this flaw is not related to Bitcoin. The protocol has not changed and was not attacked; we are talking about an attack on a &#8220;solution provider.&#8221; This calls into question this ecosystem, but not Bitcoin (a distinction of importance, as some regulators are very quick to focus on Bitcoin &#8220;because&#8221; rather than on the real problem).</p><ul><li><p><a href="https://www.youtube.com/watch?v=XuoqKYxDHVc">An interview with Elon Musk</a> - Youtube</p></li><li><p><a href="https://wizardsardine.com/blog/coldcard-rng-vulnerability/">The Coldcard flaw is bad, really bad and you should be worried</a></p></li><li><p><a href="https://itsallaboutthebit.com/memory-safety-absolutists/">Is memory safety really important?</a></p></li><li><p><a href="https://www.youtube.com/watch?v=VeAfJfxoFac">In case you have doubts, the war against Iran wasn&#8217;t about Iran and/or nuclear</a> - Youtube</p></li></ul><div><hr></div><h1>Databases</h1><p><strong>W</strong>e continue with databases, specifically Barrage, a load testing tool that allows you to test web applications as well as the database. Barrage lets you verify the performance of your application while subjecting your database (SQL and/or Redis-like, by the way) to stress tests. This is essential if you regularly face performance issues where the application seems slow, but tools like Blackfire or similar report nothing. It can help detect unused indexes, schema issues, or even cache problems, all while benefiting from a dashboard to track test progress and the impact on the stack under load.</p><p>An essential addition to your toolbox for future testing sessions.</p><ul><li><p><a href="https://github.com/singhpratech/ferrovec">Ferrovec, a WASM-powered vector store</a> - Github</p></li><li><p><a href="https://github.com/codetesla51/barrage">Barrage, a load testing tool for applications and databases</a> - Github</p></li></ul><div><hr></div><h1>Cloud Computing / Linux</h1><p><strong>A</strong> quick detour into the cloud to discover BrewFS, a &#8220;high-performance&#8221; distributed filesystem primarily marketed for AI agents and S3-type storage. You have probably already encountered performance problems when multiple agents want to access the same project/file; BrewFS offers a solution to this type of problem (and many others) and allows your agents to benefit from an FS compatible with other solutions like Redis, SQLite, and the like.</p><ul><li><p><a href="https://github.com/brewfs/brewfs">BrewFS, a high-performance distributed filesystem for AI, object storage and more</a> - Github</p></li></ul><div><hr></div><h1>Machine Learning / AI</h1><p><strong>W</strong>e return to the classics with machine learning / AI, specifically Swarm Forge, a &#8220;forge&#8221; for your agents to steer and structure their work, all built by Uncle Bob (I will refrain from giving an opinion on the man; I look at his work and appreciate what he has brought to the ecosystem, the rest is none of my business). I admit I appreciate this kind of project, especially because harness tools like Codex / Claude Code do not offer this type of granular management of your agents/sub-agents; they launch them but give very little (or no) control over them. Consequently, we often find ourselves &#8220;scolding&#8221; agents to keep them within a framework or to step out of it when necessary, and we end up wasting a lot of time because an agent &#8220;doesn&#8217;t fit into the right box.&#8221; The approach is not revolutionary, but it allows for better framing of your agents and avoids overflows that harnesses do not detect (or decide to ignore).</p><p>Another point with a YouTube video from WBD featuring Luke Gromen, who questions the viability of the AI business model and the companies driving this ecosystem. AI is probably not a bubble (at least not in the financial sense of the term), but the companies within this ecosystem are problematic. Their funding is often a closed circuit (though very dependent on the FED in the end), and they sustain themselves in a cycle of buying/selling/shares to fund the race to AGI. As a result, the ecosystem is spinning in circles, and it&#8217;s a safe bet that without the dollar, there is no guarantee it would survive.</p><ul><li><p><a href="https://github.com/unclebob/swarm-forge">SwarmForge, a tool to coordinate agents</a> - Github</p></li><li><p><a href="https://www.youtube.com/watch?v=w-Vopzt_m5g">AI is a different kind of bubble</a> - Youtube</p></li><li><p><a href="https://www.bitpanda.com/en/ai-investing-assistant">Bitpanda is opening its own MCP server</a></p></li></ul><div><hr></div><h1>PHP</h1><p><strong>W</strong>e finish with PHP, specifically a bundle that allows you to generate toast-type notifications on your applications while maintaining the HTTP cache system (if enabled). This is a problem we often encounter in our applications (after all, a notification costs nothing, so why not use them), but it hits that famous HTTP cache wall (another vast subject). The bundle uses Turbo Stream / Stimulus to bypass the block and allows you to keep your notifications in any situation.</p><p>One last topic: QueueSQL, a library for Laravel that allows you to store your write requests in queues. For Symfony lovers, this is equivalent to using a dedicated bus in Messenger with storage via RabbitMQ / Redis / Doctrine using a clean CQRS architecture (I would have a lot to say about the latter). QueueSQL also allows the use of &#8220;parallelism&#8221; (in practice, PSL and Parallel also allow this if needed).</p><ul><li><p><a href="https://symfony.com/blog/introducing-symfony-reprise-the-symfony-integration-layer-for-modern-bundlers">Symfony is unveiling Reprise, a modern layer for bundlers</a></p></li><li><p><a href="https://github.com/marilenaRM/turbo-toast-bundle">A bundle to trigger toast notification while keeping HTTP cache</a> - Github</p></li><li><p><a href="https://github.com/kamranata/queue-sql">QueueSQL, a library to queue any Laravel write through parallel batched jobs</a> - Github</p></li></ul><p>That&#8217;s it for this week&#8217;s watch. As always, feel free to leave a comment / like / share if you feel like it. For my part, I&#8217;ll see you next week for a new update.</p><p>Until then, have a great weekend and a wonderful week.</p><p>See you Friday.</p>]]></content:encoded></item><item><title><![CDATA[Weekly Tech - 27th of July 2026]]></title><description><![CDATA[WASI is gaining more functionality, open-weight models are starting to put pressure on private AI companies, and PHP is (re)turning to C to boost efficiency... ready?]]></description><link>https://guikingone.substack.com/p/weekly-tech-27th-of-july-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/weekly-tech-27th-of-july-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 31 Jul 2026 14:02:33 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/209243137/01057a32ada3938682df3c7e17512061.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>H</strong>ello everyone, I hope you&#8217;re all doing well?</p><p>New week, new Weekly Tech, and this week is intended to be much lighter than usual. Indeed, some topics took longer than expected and I ran out of time during the week to read certain articles (these will be postponed to next week), so I focused on machine learning and PHP to keep it readable. I&#8217;ll have the opportunity to return to other subjects later.</p><p>Anyway, enough talking, shall we dive in?</p><div><hr></div><h1>General Knowledge</h1><p><strong>W</strong>e start with general knowledge and a video in French about Marx. I know it&#8217;s economics and not necessarily development, but it&#8217;s related, especially because the notion of value is a concept that comes up often in the mouths of certain economists, even more so today in the age of AI. I wanted to cite this video to remind us that Marx was wrong, and that it is time to admit it so that everyone understands that these theories were outdated (and already were in his time) and that continuing to base oneself on his ideas in 2026 stems from a misunderstanding of economics and what underpins our societies.</p><ul><li><p><a href="https://www.youtube.com/watch?v=bbv0G8Oy2lI">Marx was wrong, deal with that</a> - Youtube (French)</p></li></ul><div><hr></div><h1>Cloud computing / Linux</h1><p><strong>W</strong>e&#8217;re back to the classics with cloud, specifically a video explaining how to add a screen to a Raspberry Pi 5, which is very useful if you use it to build dashboards, manage home automation, or even monitor applications remotely (hi Claude).</p><p>Another topic with WASIX, an extension of the WASI standard that now supports threads and other goodies. I&#8217;m slowly getting back into WASM / WASI and I must admit that the lack of support for threads (and therefore potentially &#8220;parallelism&#8221; for certain tasks) had pushed me away at the time. I don&#8217;t know where WASIX will go, but the first drafts provide hope.</p><ul><li><p><a href="https://www.youtube.com/watch?v=34D1imLordU">Adding a screen to a Raspberry Pi 5</a> - Youtube</p></li><li><p><a href="https://wasix.org/">Wasix, a superset on top of WASI</a></p></li></ul><div><hr></div><h1>Machine learning / AI</h1><p><strong>R</strong>ight, what&#8217;s new on the machine learning side? Quite a few topics this week, notably with the definitive open-weight release of Kimi K3 (it was restricted to an API until its release on Huggingface). The model has caused quite a stir (I haven&#8217;t had the chance to test it yet since I use Ollama, and they decided to switch it to Pay-as-You-Go&#8212;even on Max subscriptions&#8212;) and seems to be rapidly approaching frontier models like Fable (although Fable &#8220;unofficially&#8221; came out at the beginning of the year, meaning Kimi has a larger lag than expected). I&#8217;m very curious to see what this will bring in the coming months.</p><p>Another point with an article on the real risk within the AI ecosystem. I largely agree with the article that the risk does not lie so much in open-weight / open-source models (in practice, these models can be verified, run locally, and easily audited). I believe the risk actually lies in models developed by the likes of OpenAI / Anthropic, since no one has control over what they do and performing an audit is nearly impossible without relying on their own claims. However, I&#8217;m not asking for more regulation; but seeing companies like OpenAI / Anthropic open their models and make them available (at least as open-weight) would be a huge plus for everyone.</p><p><em>Useful reminder that the Huggingface incident required OpenAI to verify its own infrastructure before finding the path used during the attack.</em></p><p>The article also echoes another piece from Anthropic calling for &#8220;more controls&#8221; on these open-weight models. This amounts to imposing more regulation and thus more state control over innovations. I think this is a mistake that will be very costly if governments less &#8220;friendly&#8221; toward innovation are elected and end up &#8220;censoring just in case&#8221; (hello Trump administration and Mythos).</p><p>Final article by the Cursor team and the use of a fleet of agents (a &#8220;swarm of agents&#8221; according to their own terminology) to trigger the development of a SQLite clone via different models. The metrics speak for themselves, and I think the agent swarm is the solution for the future (especially since centralizing everything in a single agent is not viable in terms of context and execution speed). We&#8217;ll see where this leads.</p><ul><li><p><a href="https://github.com/rynfar/meridian">Meridian, use your Claude Max subscription with other harness</a> - Github</p></li><li><p><a href="https://blogs.nvidia.com/blog/open-secure-ai-alliance/">The Open Secure AI Alliance is created, without OpenAI and Anthropic</a></p></li><li><p><a href="https://huggingface.co/moonshotai/Kimi-K3">Kimi K3 is now open-weight</a> - Huggingface</p></li><li><p><a href="https://antirez.com/news/172">The real AI risk is inside closed labs</a></p></li><li><p><a href="https://github.com/DeusData/codebase-memory-mcp">CodebaseMemoryMCP, a high performance code intelligence MCP server</a> - Github</p></li><li><p><a href="https://github.com/spelcc/gate">Gate, a local MCP gateway</a> - Github</p></li><li><p><a href="https://www.anthropic.com/news/position-open-weights-models">Anthropic is pushing for stricter regulations against open-weight models</a></p></li><li><p><a href="https://github.com/qaml-ai/camelAI">CamelAI, an AI agent built for Cloudflare Workers and Durable Objects</a> - Github</p></li><li><p><a href="https://cursor.com/fr/blog/agent-swarm-model-economics">Could models perform better when triggered as a team?</a></p></li></ul><div><hr></div><h1>PHP</h1><p><strong>W</strong>e finish with PHP and specifically Turbo, an extension for PHPStan. The idea is to make it more efficient by using a C extension (which, when you think about it, amounts to rewriting PHPStan in C). Since PHPStan is limited by the PHP parser, performance is impacted (and having used it, there isn&#8217;t really a solution other than using cache and &#8220;praying&#8221; that everything goes well), and we end up having to manage &#8220;pseudo-parallelism&#8221; to compensate for the problems induced by the parser.</p><p>Another point (indirectly related) regarding the rewriting of the PHP virtual machine in Rust. The project was launched by the developers at Jolicode and opens new doors for PHP. As a reminder, the PHP Foundation had discussed the idea of introducing Rust into PHP to optimize / modernize part of its code. I&#8217;m very curious to see where this project goes.</p><ul><li><p><a href="https://github.com/phpstan/turbo-ext">Turbo, a PHPStan extension</a> - Github</p></li><li><p><a href="https://github.com/EncoreDigitalGroup/heimdall">Heimdall, a Composer plugin to protect against supply chain attacks</a> - Github</p></li><li><p><a href="https://blog.jetbrains.com/phpstorm/2026/07/key-takeaways-from-phpverse-2026/">The key takeaways from PHPVerse 2026</a></p></li><li><p><a href="https://jolicode.com/blog/writing-the-php-virtual-machine-in-rust-with-a-lot-of-help-from-ai">(Re)writing the PHP Virtual machine in Rust</a></p></li></ul><p>That&#8217;s all for this Weekly Tech. As usual, feel free to leave a comment / like / share if you feel like it. For my part, I&#8217;ll see you next week for a new update.</p><p>Until then, have a great weekend and a wonderful week.</p><p>See you Friday.</p>]]></content:encoded></item><item><title><![CDATA[Veille de la semaine du 27 Juillet 2026]]></title><description><![CDATA[WASI gagne en fonctionnalit&#233;, l'open-weight commence &#224; peser sur les entreprises priv&#233;es de l'IA, PHP se (re)met &#224; C pour gagner en efficience, pr&#234;t(e)s ?]]></description><link>https://guikingone.substack.com/p/veille-de-la-semaine-du-27-juillet</link><guid isPermaLink="false">https://guikingone.substack.com/p/veille-de-la-semaine-du-27-juillet</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 31 Jul 2026 14:01:42 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/209242649/4138b911f33082e99b6643850f114aa4.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>B</strong>ien le bonjour &#224; tous et &#224; toutes, j&#8217;esp&#232;re que tout le monde va bien ?</p><p>Nouvelle semaine, nouvelle veille et cette semaine elle se veut beaucoup plus l&#233;g&#232;re que d&#8217;habitude, en effet, certains sujets ont pris plus de temps que pr&#233;vu et j&#8217;ai manqu&#233; de temps dans la semaine pour lire certains articles (ce sera donc d&#233;cal&#233; &#224; la semaine prochaine), je me suis donc concentr&#233; sur le machine learning et sur PHP afin de rester lisible, j&#8217;aurais l&#8217;occasion de revenir sur d&#8217;autres sujets plus tard.</p><p>Bref, assez parl&#233;, on se lance ?</p><div><hr></div><h1>Culture g&#233;n&#233;rale</h1><p><strong>O</strong>n commence donc par la culture g&#233;n&#233;rale et une vid&#233;o en fran&#231;ais sur Marx, je sais c&#8217;est de l&#8217;&#233;conomie et pas forc&#233;ment du d&#233;veloppement mais c&#8217;est li&#233;, notamment parce que la notion de valeur est une notion qui revient beaucoup dans la bouche de certains &#233;conomistes, encore plus aujourd&#8217;hui &#224; l&#8217;&#232;re de l&#8217;IA. Je tenais donc &#224; citer cette vid&#233;o afin de rappeler que Marx avait tort, qu&#8217;il est temps de l&#8217;admettre afin que tout le monde comprenne bien que ces th&#233;ories &#233;taient d&#233;pass&#233;es (et l&#8217;&#233;taient d&#233;j&#224; &#224; son &#233;poque) et que continuer &#224; se baser sur ses id&#233;es en 2026 rel&#232;ve d&#8217;une incompr&#233;hension de l&#8217;&#233;conomie et de ce qui sous-tend nos soci&#233;t&#233;s.</p><ul><li><p><a href="https://www.youtube.com/watch?v=bbv0G8Oy2lI">Marx was wrong, deal with that</a> - Youtube (French)</p></li></ul><div><hr></div><h1>Cloud computing / Linux</h1><p><strong>O</strong>n revient au classique avec le cloud et notamment une vid&#233;o expliquant comment ajouter un &#233;cran sur un Raspberry Pi 5, tr&#232;s utile si vous utilisez ce dernier pour construire des dashboard, g&#233;rer de la domotique voir suivre des applications &#224; distance (coucou Claude).</p><p>Autre sujet avec WASIX, une extension du standard WASI qui supporte cette fois-ci les threads et autres joyeuset&#233;s, je me remet doucement sur WASM / WASI et je dois avouer que le manque de support pour les threads (et donc potentiellement un &#8220;parall&#233;lisme&#8221; de certaines t&#226;ches) m&#8217;avait mis &#224; l&#8217;&#233;cart &#224; l&#8217;&#233;poque, je ne sais pas o&#249; ira WASIX mais les premi&#232;res &#233;bauches donnent de l&#8217;espoir.</p><ul><li><p><a href="https://www.youtube.com/watch?v=34D1imLordU">Adding a screen to a Raspberry Pi 5</a> - Youtube</p></li><li><p><a href="https://wasix.org/">Wasix, a superset on top of WASI</a></p></li></ul><div><hr></div><h1>Machine learning / AI</h1><p><strong>B</strong>ien, quoi de neuf du c&#244;t&#233; du machine learning ? Pas mal de sujets cette semaine, notamment avec la sortie de Kimi K3 en open-weight de fa&#231;on d&#233;finitive (il &#233;tait cantonn&#233; &#224; une API jusqu&#8217;&#224; sa sortie sur Huggingface), le mod&#232;le a fait pas mal de bruit (je n&#8217;ai pas encore eu l&#8217;occasion de le tester puisque je passe par Ollama et ces derniers ont d&#233;cid&#233; de le passer en Pay-as-You-Go - et ce, m&#234;me sur des abonnements Max -) et semble se rapprocher rapidement des mod&#232;les frontiers type Fable (m&#234;me si Fable est &#8220;officieusement&#8221; sorti en d&#233;but d&#8217;ann&#233;e, ce qui laisse donc un retard plus important que pr&#233;vu pour Kimi), tr&#232;s curieux de voir ce que cela va apporter dans les mois &#224; venir.</p><p>Autre point avec un article sur le risque r&#233;el au sein de l&#8217;&#233;cosyst&#232;me IA, je suis assez d&#8217;accord avec l&#8217;article sur le fait que le risque ne se situe pas tant dans les mod&#232;les open weight / open source (en pratique ces mod&#232;les-l&#224; peuvent &#234;tre v&#233;rifi&#233;s, lanc&#233;s localement, audit&#233;s facilement), je pense effectivement que le risque se situe dans les mod&#232;les qui sont d&#233;velopp&#233;s par des OpenAI / Anthropic puisque personne n&#8217;a la main sur ce qu&#8217;ils font et qu&#8217;effectuer un audit est quasiment impossible sans se baser sur ce qu&#8217;ils avancent comme argument, attention cependant, je ne demande pas plus de r&#233;gulations mais voir des entreprises comme OpenAI / Anthropic ouvrir leur mod&#232;les et les rendre disponibles (au moins en open-weight) serait un &#233;norme plus pour tout le monde.</p><p><em>Rappel utile que l&#8217;incident avec Hugginface a n&#233;cessit&#233; qu&#8217;OpenAI aille v&#233;rifier ses propres infrastructures avant de trouver le cheminement utilis&#233; durant l&#8217;attaque.</em></p><p>L&#8217;article fait d&#8217;ailleurs &#233;cho &#224; un autre article venant d&#8217;Anthropic et appelant &#224; &#8220;plus de contr&#244;les&#8221; sur ces mod&#232;les open-weight, cela revient &#224; imposer plus de r&#233;gulations et donc plus de contr&#244;les de l&#8217;Etat sur des innovations, je pense que c&#8217;est une erreur qui co&#251;tera tr&#232;s cher si des gouvernements moins &#8220;amicaux&#8221; envers l&#8217;innovation sont &#233;lus et finissent par &#8220;censurer parce qu&#8217;on ne sait jamais&#8221; (coucou l&#8217;administration Trump et Mythos).</p><p>Dernier article par l&#8217;&#233;quipe de Cursor et le fait d&#8217;avoir utilis&#233; une flotte d&#8217;agents (un essaim d&#8217;agents selon leur propres terminologies) pour d&#233;clencher le d&#233;veloppement d&#8217;un clone de SQLite via diff&#233;rents mod&#232;les, les m&#233;triques parlent d&#8217;elles-m&#234;mes et je pense que la flotte d&#8217;agents est une solution pour l&#8217;avenir (notamment car tout centraliser en un seul agent n&#8217;est pas viable en terme de contexte et de rapidit&#233; d&#8217;ex&#233;cution), &#224; voir o&#249; cela nous m&#232;nera.</p><ul><li><p><a href="https://github.com/rynfar/meridian">Meridian, use your Claude Max subscription with other harness</a> - Github</p></li><li><p><a href="https://blogs.nvidia.com/blog/open-secure-ai-alliance/">The Open Secure AI Alliance is created, without OpenAI and Anthropic</a></p></li><li><p><a href="https://huggingface.co/moonshotai/Kimi-K3">Kimi K3 is now open-weight</a> - Huggingface</p></li><li><p><a href="https://antirez.com/news/172">The real AI risk is inside closed labs</a></p></li><li><p><a href="https://github.com/DeusData/codebase-memory-mcp">CodebaseMemoryMCP, a high performance code intelligence MCP server</a> - Github</p></li><li><p><a href="https://github.com/spelcc/gate">Gate, a local MCP gateway</a> - Github</p></li><li><p><a href="https://www.anthropic.com/news/position-open-weights-models">Anthropic is pushing for stricter regulations against open-weight models</a></p></li><li><p><a href="https://github.com/qaml-ai/camelAI">CamelAI, an AI agent built for Cloudflare Workers and Durable Objects</a> - Github</p></li><li><p><a href="https://cursor.com/fr/blog/agent-swarm-model-economics">Could models perform better when triggered as a team?</a></p></li></ul><div><hr></div><h1>PHP</h1><p><strong>O</strong>n finit avec PHP et notamment Turbo, une extension pour PHPStan, l&#8217;id&#233;e &#233;tant de rendre ce dernier plus efficace en passant par une extension en C (cette approche revenant &#224; r&#233;&#233;crire PHPStan en C quand on y pense), PHPStan &#233;tant limit&#233; par le parser PHP, les performances s&#8217;en trouvent impact&#233;es (et pour l&#8217;avoir pratiqu&#233;, il n&#8217;y a pas r&#233;ellement de solution en dehors d&#8217;utiliser du cache et &#8220;prier&#8221; que tout se passe bien) et on se retrouve &#224; devoir g&#233;rer un &#8220;pseudo-parall&#233;lisme&#8221; pour compenser les probl&#232;mes induits par le parser.</p><p>Autre point (li&#233; indirectement) avec le fait de r&#233;&#233;crire la machine virtuelle PHP en Rust, le projet est lanc&#233; par les d&#233;veloppeurs de Jolicode et ouvre de nouvelles portes &#224; PHP, pour rappel, la PHP Foundation avait abord&#233; l&#8217;id&#233;e d&#8217;introduire Rust dans PHP afin d&#8217;optimiser / moderniser une partie de son code, tr&#232;s curieux de voir o&#249; le projet ira en tout cas.</p><ul><li><p><a href="https://github.com/phpstan/turbo-ext">Turbo, a PHPStan extension</a> - Github</p></li><li><p><a href="https://github.com/EncoreDigitalGroup/heimdall">Heimdall, a Composer plugin to protect against supply chain attacks</a> - Github</p></li><li><p><a href="https://blog.jetbrains.com/phpstorm/2026/07/key-takeaways-from-phpverse-2026/">The key takeaways from PHPVerse 2026</a></p></li><li><p><a href="https://jolicode.com/blog/writing-the-php-virtual-machine-in-rust-with-a-lot-of-help-from-ai">(Re)writing the PHP Virtual machine in Rust</a></p></li></ul><p>Voici donc pour la veille de cette semaine, comme d&#8217;habitude, n&#8217;h&#233;sitez pas &#224; laisser un commentaire / like / partager si le coeur vous en dit, de mon c&#244;t&#233;, je vous donne rendez-vous d&#232;s la semaine prochaine pour une nouvelle veille.</p><p>D&#8217;ici-l&#224;, passez un excellent weekend ainsi qu&#8217;une tr&#232;s bonne semaine.</p><p>A vendredi.</p>]]></content:encoded></item><item><title><![CDATA[Veille de la semaine du 20 Juillet 2026]]></title><description><![CDATA[La surveillance de masse s'installe, l'IA se met &#224; attaquer ses "semblables", PHP ouvre la porte aux mod&#232;les locaux pour la voix et Wasm revient sous les projecteurs, pr&#234;t(e)s ?]]></description><link>https://guikingone.substack.com/p/veille-de-la-semaine-du-20-juillet</link><guid isPermaLink="false">https://guikingone.substack.com/p/veille-de-la-semaine-du-20-juillet</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 24 Jul 2026 14:31:42 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/208336705/6aedfcc4d5734be5935eb1be364bac0d.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>B</strong>ien le bonjour &#224; tous et &#224; toutes, j&#8217;esp&#232;re que tout le monde va bien.</p><p>Nouvelle semaine, nouvelle veille et cette semaine, peu de gros sujets, j&#8217;avoue avoir &#233;t&#233; en d&#233;placement cette semaine et avoir manqu&#233; de temps pour traiter certains sujets, ma liste de lecture sur X commence &#224; devenir plus que cons&#233;quente et je ne pr&#233;f&#232;re pas parler de ma playlist Youtube. Bref, les sujets arrivent au fur et &#224; mesure avec un peu de retard, j&#8217;ai fait glisser sur cette semaine les sujets qui m&#8217;ont le plus marqu&#233; et que j&#8217;avais d&#233;j&#224; pu lire en fin de semaine derni&#232;re.</p><p>Bref assez parl&#233;, on se lance ?</p><div><hr></div><h1>Culture g&#233;n&#233;rale</h1><p><strong>O</strong>n commence donc par la culture g&#233;n&#233;rale, avec un premier sujet, qui concerne surtout les Europ&#233;ens avec le vote sur Chat Control (que vous avez peut-&#234;tre suivi), si vous n&#8217;avez pas suivi, l&#8217;id&#233;e est que Chat Control va ouvrir la possibilit&#233; de cr&#233;er une &#8220;surveillance de masse&#8221; sur les syst&#232;mes de messagerie, c&#8217;est (&#224; mes yeux) un viol de la libert&#233; d&#8217;expression (&#224; partir du moment o&#249; vous vous savez surveill&#233;s, vous vous censurez naturellement) et de la vie priv&#233;e, le &#8220;syst&#232;me&#8221; se d&#233;fend en mettant en avant la d&#233;fense des mineurs (ce qui est louable mais le diable se cache dans les d&#233;tails) mais on sent poindre certains sujets moins r&#233;jouissants notamment le fait de surveiller vos messages, de les stocker pour les analyser et donc, &#224; terme, de pouvoir retracer une partie de vos &#233;changes, le tout sans votre consentement. Des solutions existents telles que Signal ou via des solutions d&#233;centralis&#233;es (j&#8217;en ai d&#233;j&#224; parl&#233; il y a quelques temps) mais j&#8217;avoue que voir ce genre de votes passer sans m&#234;me consulter le peuple me d&#233;range (je ne suis pas un grand partisan de l&#8217;U.E, je pr&#233;f&#232;re &#234;tre franc &#224; ce sujet, ce vote n&#8217;arrange rien), surtout quand on met en avant &#8220;la d&#233;fense des innocents&#8221; sans prendre en compte (ou en les prenant et les excluant, l&#224; serait le pire) les effets secondaires.</p><p>Autre sujet sur la gestion de la monnaie du c&#244;t&#233; de la France, je sais que c&#8217;est un sujet qui n&#8217;est pas forc&#233;ment lu par tout le monde mais ce dernier reste important (et li&#233; indirectement au pr&#233;c&#233;dent d&#8217;ailleurs), la vid&#233;o revient sur les d&#233;cisions prises quant &#224; la gestion de la monnaie durant les p&#233;riodes post-R&#233;volution et l&#8217;impact de ce type de gestion &#8220;au doigt mouill&#233;&#8221; sur les citoyens et les d&#233;cisions que ces derniers prennent (pensez toujours au fait que quand une d&#233;cision est prise, elle est motiv&#233;e de pr&#232;s comme de loin par la monnaie sous-jacente).</p><ul><li><p><a href="https://www.youtube.com/watch?v=nRqqihy7l10">Chat Control is a massive problem, even if you think you&#8217;re safe</a> - Youtube</p></li><li><p><a href="https://www.youtube.com/watch?v=bcFCf3q8y9E">Renault built an &#8220;electric mini supercar&#8221;, the Renault 5 Turbo 3E</a> - Youtube</p></li><li><p><a href="https://github.com/solal3105/nadir2">Want to cool your room? Here&#8217;s Nadir, a tracker for your windows</a> - Github</p></li><li><p><a href="https://www.youtube.com/watch?v=lzctJuoJTgA">France destroyed its own money, due to its own printing of it</a> - Youtube (French)</p></li></ul><div><hr></div><h1>Cloud computing / Linux</h1><p><strong>T</strong>out petit d&#233;tour du c&#244;t&#233; du cloud avec Glide qui est un sujet qui a gliss&#233; dans la veille il y a maintenant un mois et demi, je n&#8217;avais pas eu le temps d&#8217;essayer ou du moins pas le temps d&#8217;explorer les diff&#233;rents clients, si vous avez louper l&#8217;information, Glide est une s&#233;rie de clients pour Valkey, une alternative open-source &#224; Redis sur diff&#233;rents languages et avec un support plus complet des fonctionnalit&#233;s que le protocol natif de Redis.</p><p>Dernier sujet avec Nectar qui est un nouveau langage de programmation qui compile en WebAssembly, j&#8217;avoue ne pas saisir l&#8217;int&#233;r&#234;t imm&#233;diat de cr&#233;er un nouveau language pour cela (surtout quand Javascript le permet d&#233;j&#224; indirectement) mais soit.</p><ul><li><p><a href="https://github.com/valkey-io/valkey-glide">Glide, a set of clients for Valkey</a> - Github</p></li><li><p><a href="https://github.com/Neirda24/envoke">Envoke, a cross-platform directory enter/leave hooks</a> - Github</p></li><li><p><a href="https://github.com/HibiscusConsulting/nectar-lang">Nectar, a programming language that compiles to WebAssembly</a> - Github</p></li></ul><div><hr></div><h1>Database</h1><p><strong>T</strong>r&#232;s l&#233;ger d&#233;tour du c&#244;t&#233; de la base de donn&#233;es pour d&#233;couvrir FlareDB, une surcouche &#224; Apache Beam, une BDD orient&#233;e streaming, j&#8217;ai pas eu le temps de la tester enti&#232;rement donc je vous laisserai y jeter un oeil, j&#8217;aurai l&#8217;occasion de revenir dessus dans les semaines qui viennent si mes tests apportent des arguments pr&#233;cis.</p><ul><li><p><a href="https://github.com/flare-db/flare-db">FlareDB, a streaming database built on top for Apache Beam</a> - Github</p></li></ul><div><hr></div><h1>Machine learning / AI</h1><p><strong>B</strong>ien, quoi de neuf du c&#244;t&#233; du machine learning ? La semaine a &#233;t&#233; relativement calme, pas de grosses sorties de mod&#232;les en dehors de Qwen qui sort sa version 3.0 pour le TTS, une grosse partie de l&#8217;&#233;cosyst&#232;me attend Opus 5 aujourd&#8217;hui ou semaine prochaine (je parierais plus sur la semaine prochaine).</p><p>Autre sujet avec Colibri, un moteur d&#8217;ex&#233;cution pour GLM.5.2 &#233;crit en C, l&#8217;id&#233;e &#233;tant de pouvoir faire rentrer le mod&#232;le &#8220;complet&#8221; sur une carte graphique type RTX 4090, des tests sont en cours de mon c&#244;t&#233;, j&#8217;aurais probablement des retours &#224; formuler d&#232;s la semaine prochaine.</p><p>Autre sujet (qui a fait un peu plus de bruit pour le coup) le fait que Huggingface a &#233;t&#233; attaqu&#233; via un mod&#232;le, Huggingface pr&#233;cisait que le mod&#232;le en question m&#234;me s&#8217;il n&#8217;&#233;tait pas connu, agissait seul et semblait prendre des d&#233;cisions relativement seul sur la fa&#231;on dont l&#8217;attaque &#233;tait men&#233;e et comment rediriger certains appels au &#8220;bon moment&#8221;, le temps a fait son oeuvre et force est de constater que le mod&#232;le qui a attaqu&#233; Hunggingface est un GPT 5.6 (aid&#233; potentiellement d&#8217;un mod&#232;le interne &#224; GPT qui n&#8217;a pas encore &#233;t&#233; publi&#233;, on peut donc penser potentiellement &#224; un 5.7), un post-mortem a &#233;t&#233; publi&#233; par openAI (Anthropic gagnerait &#224; s&#8217;inspirer de cette approche plut&#244;t que de hurler au loup) et indique comment l&#8217;attaque a &#233;t&#233; men&#233;e, tr&#232;s int&#233;ressant, je vous recommande la lecture.</p><ul><li><p><a href="https://funaudiollm.github.io/qwen-audio-3.0-tts/">Qwen 3.0 TTS is out</a></p></li><li><p><a href="https://github.com/Osmantic/ODS">ODS, turn your mac / Windows / Linux into a private AI server</a> - Github</p></li><li><p><a href="https://github.com/JustVugg/colibri">Colibri, a tiny Engine to run GLM 5.2 on machine with ~ 25Gb RAM</a> - Github</p></li><li><p><a href="https://github.com/promptfoo/promptfoo">Promptfoo, a CLI to evaluate and red-teaming LLM applications</a> - Github</p></li><li><p><a href="https://thealliance.ai/projects/tapestry">Tapestry, a consortium approach to train frontier models</a></p></li><li><p><a href="https://les-tilleuls.coop/blog/un-rag-magique">Building a RAG around &#8220;Magic: The Gathering&#8221;</a> (French)</p></li><li><p><a href="https://github.com/ayghri/i-have-adhd">An ADHD friendly skill for Claude Code / Codex</a> - Github</p></li><li><p><a href="https://openai.com/index/hugging-face-model-evaluation-security-incident/">Looks like GPT 5.6 Sol was used in the attack against Huggingface</a></p></li></ul><div><hr></div><h1>PHP</h1><p><strong>T</strong>out petit d&#233;tour du c&#244;t&#233; de PHP pour d&#233;couvrir Vocalizer, une librairie supportant le TTS via les mod&#232;les locaux, c&#8217;est un &#233;quivalent de Voicebox (j&#8217;en ai parl&#233; il y a quelques temps dans la veille, Voicebox permet de copier des voix via une GUI). L&#8217;id&#233;e ici est de le faire directement depuis PHP via une librairie directe qui va r&#233;cup&#233;rer les mod&#232;les en local et vous autoriser &#224; les configurer, la librairie ressemble &#233;norm&#233;ment &#224; TransformersPHP, qui utilisait les mod&#232;les depuis Huggingface</p><p>Autre sujet avec l&#8217;utilisation de Vektor afin de stocker et permettre la recherche s&#233;mantique sur la documentation PHP, l&#8217;utilisation est triviale (et n&#8217;a que peu d&#8217;impact sur l&#8217;espace disque) et le r&#233;sultat semble plut&#244;t coh&#233;rent, une tr&#232;s bonne alternative aux solutions n&#233;cessitant des serveurs d&#233;di&#233;s.</p><ul><li><p><a href="https://github.com/AkramZerarka/vocalizer">Vocalizer, a TTS library for PHP</a> - Github</p></li><li><p><a href="https://www.exakat.io/using-a-vector-database-with-php">Using a vector database with PHP</a></p></li></ul><div><hr></div><h1>Rust / Go</h1><p><strong>D</strong>ernier sujet avec Type/ Rust et la compilation d&#8217;un parser Typescript vers WASM, ce dernier commence &#224; reprendre un petit peu son souffle, j&#8217;avoue ne pas avoir totalement suivi l&#8217;&#233;cosyst&#232;me ces derni&#232;res ann&#233;es, toujours int&#233;ressant de voir des initiatives de ce type voir le jour.</p><ul><li><p><a href="https://encore.dev/blog/typescript-parser-wasm">Compiling a Typescript parser to WASM</a></p></li></ul><p>Voici donc pour la veille de cette semaine, comme d&#8217;habitude, n&#8217;h&#233;sitez pas &#224; laisser un commentaire / like / partager si le coeur vous en dit, de mon c&#244;t&#233;, je vous donne rendez-vous d&#232;s la semaine prochaine pour une nouvelle veille.</p><p>D&#8217;ici-l&#224;, passez un excellent weekend ainsi qu&#8217;une tr&#232;s bonne semaine.</p><p>A vendredi.</p>]]></content:encoded></item><item><title><![CDATA[Weekly Tech - 20th of July 2026]]></title><description><![CDATA[Mass surveillance is taking hold, AI is starting to attack its 'peers,' PHP opens the door to local voice models, and Wasm is back in the spotlight... ready?]]></description><link>https://guikingone.substack.com/p/weekly-tech-20th-of-july-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/weekly-tech-20th-of-july-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 24 Jul 2026 14:31:40 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/208337501/020c614b27ce8ed23c50f9d8d6c41575.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>H</strong>ello everyone, I hope you are all doing well.</p><p>New week, new Weekly Tech, and this week there are few major topics. I must admit I was traveling this week and lacked the time to cover certain subjects; my reading list on X is becoming quite substantial, and I&#8217;d rather not even mention my YouTube playlist. In short, the topics are arriving gradually and with a bit of a delay; I&#8217;ve pushed over to this week the subjects that struck me the most and which I had already been able to read late last week.</p><p>Anyway, enough talking, shall we dive in?</p><div><hr></div><h1>General Knowledge</h1><p><strong>W</strong>e start with general knowledge, with a first topic mainly concerning Europeans regarding the vote on Chat Control (which you may have followed). If you haven&#8217;t, the idea is that Chat Control will open the possibility of creating &#8220;mass surveillance&#8221; on messaging systems. In my eyes, this is a violation of freedom of expression (from the moment you know you are being watched, you naturally censor yourself) and privacy. The &#8220;system&#8221; defends itself by highlighting the protection of minors (which is laudable, but the devil is in the details), but we can see some less pleasant subjects emerging, notably the fact of monitoring your messages, storing them for analysis, and therefore, eventually, being able to trace part of your exchanges&#8212;all without your consent. Solutions exist, such as Signal or decentralized options (which I mentioned a while ago), but I admit that seeing these kinds of votes pass without even consulting the people bothers me (I am not a great supporter of the E.U., I prefer to be honest about this, and this vote doesn&#8217;t help), especially when &#8220;defending the innocent&#8221; is put forward without considering (or by considering them and excluding them, which would be the worst) the side effects.</p><p>Another topic regarding currency management in France; I know this is a subject that isn&#8217;t necessarily read by everyone, but it remains important (and indirectly linked to the previous one). The video reviews the decisions made regarding currency management during the post-Revolution periods and the impact of this kind of &#8220;rule-of-thumb&#8221; management on citizens and the decisions they make (always remember that when a decision is made, it is motivated, directly or indirectly, by the underlying currency).</p><ul><li><p><a href="https://www.youtube.com/watch?v=nRqqihy7l10">Chat Control is a massive problem, even if you think you&#8217;re safe</a> - Youtube</p></li><li><p><a href="https://www.youtube.com/watch?v=bcFCf3q8y9E">Renault built an &#8220;electric mini supercar&#8221;, the Renault 5 Turbo 3E</a> - Youtube</p></li><li><p><a href="https://github.com/solal3105/nadir2">Want to cool your room? Here&#8217;s Nadir, a tracker for your windows</a> - Github</p></li><li><p><a href="https://www.youtube.com/watch?v=lzctJuoJTgA">France destroyed its own money, due to its own printing of it</a> - Youtube (French)</p></li></ul><div><hr></div><h1>Cloud computing / Linux</h1><p><strong>A</strong> quick detour to the cloud with Glide, a topic that slipped into the watch about a month and a half ago. I hadn&#8217;t had time to try it or at least explore the different clients. If you missed the information, Glide is a set of clients for Valkey, an open-source alternative to Redis, available in various languages and with more complete feature support than the native Redis protocol.</p><p>Last topic with Nektar, which is a new programming language that compiles to WebAssembly. I admit I don&#8217;t grasp the immediate interest in creating a new language for this (especially since Javascript already allows it indirectly), but regardless.</p><ul><li><p><a href="https://github.com/valkey-io/valkey-glide">Glide, a set of clients for Valkey</a> - Github</p></li><li><p><a href="https://github.com/Neirda24/envoke">Envoke, a cross-platform directory enter/leave hooks</a> - Github</p></li><li><p><a href="https://github.com/HibiscusConsulting/nectar-lang">Nectar, a programming language that compiles to WebAssembly</a> - Github</p></li></ul><div><hr></div><h1>Database</h1><p><strong>A</strong> very brief detour to databases to discover FlareDB, a layer on top of Apache Beam, a streaming-oriented DB. I haven&#8217;t had time to test it fully, so I&#8217;ll let you take a look; I&#8217;ll have the opportunity to come back to it in the coming weeks if my tests provide specific arguments.</p><ul><li><p><a href="https://github.com/flare-db/flare-db">FlareDB, a streaming database built on top for Apache Beam</a> - Github</p></li></ul><div><hr></div><h1>Machine learning / AI</h1><p><strong>W</strong>ell, what&#8217;s new in machine learning? The week has been relatively quiet, with no major model releases aside from Qwen releasing version 3.0 for TTS. A large part of the ecosystem is waiting for Opus 5 today or next week (I would bet more on next week).</p><p>Another topic is Colibri, an execution engine for GLM.5.2 written in C, the idea being to fit the &#8220;complete&#8221; model onto a graphics card like the RTX 4090. Tests are underway on my end; I will likely have feedback to share as early as next week.</p><p>Another subject (which made a bit more noise this time) is the fact that Huggingface was attacked via a model. Huggingface specified that the model in question, even if it was unknown, acted alone and seemed to make decisions relatively independently on how the attack was conducted and how to redirect certain calls at the &#8220;right time.&#8221; Time has done its work and it turns out that the model that attacked Huggingface is a GPT 5.6 (potentially aided by an internal GPT model that hasn&#8217;t been published yet, so we could potentially think of a 5.7). A post-mortem was published by OpenAI (Anthropic would do well to be inspired by this approach rather than crying wolf) and indicates how the attack was carried out. Very interesting, I recommend reading it.</p><ul><li><p><a href="https://funaudiollm.github.io/qwen-audio-3.0-tts/">Qwen 3.0 TTS is out</a></p></li><li><p><a href="https://github.com/Osmantic/ODS">ODS, turn your mac / Windows / Linux into a private AI server</a> - Github</p></li><li><p><a href="https://github.com/JustVugg/colibri">Colibri, a tiny Engine to run GLM 5.2 on machine with ~ 25Gb RAM</a> - Github</p></li><li><p><a href="https://github.com/promptfoo/promptfoo">Promptfoo, a CLI to evaluate and red-teaming LLM applications</a> - Github</p></li><li><p><a href="https://thealliance.ai/projects/tapestry">Tapestry, a consortium approach to train frontier models</a></p></li><li><p><a href="https://les-tilleuls.coop/blog/un-rag-magique">Building a RAG around &#8220;Magic: The Gathering&#8221;</a> (French)</p></li><li><p><a href="https://github.com/ayghri/i-have-adhd">An ADHD friendly skill for Claude Code / Codex</a> - Github</p></li><li><p><a href="https://openai.com/index/hugging-face-model-evaluation-security-incident/">Looks like GPT 5.6 Sol was used in the attack against Huggingface</a></p></li></ul><div><hr></div><h1>PHP</h1><p><strong>A</strong> short detour to PHP to discover Vocalizer, a library supporting TTS via local models. It&#8217;s an equivalent to Voicebox (I mentioned it a while ago in the watch; Voicebox allows copying voices via a GUI). The idea here is to do it directly from PHP via a direct library that retrieves models locally and allows you to configure them. The library closely resembles TransformersPHP, which used models from Huggingface.</p><p>Another topic is the use of Vektor to store and enable semantic search on PHP documentation. The use is trivial (and has little impact on disk space) and the result seems quite coherent&#8212;a very good alternative to solutions requiring dedicated servers.</p><ul><li><p><a href="https://github.com/AkramZerarka/vocalizer">Vocalizer, a TTS library for PHP</a> - Github</p></li><li><p><a href="https://www.exakat.io/using-a-vector-database-with-php">Using a vector database with PHP</a></p></li></ul><div><hr></div><h1>Rust / Go</h1><p><strong>L</strong>ast topic with Type/ Rust and the compilation of a Typescript parser to WASM. This one is starting to regain a bit of momentum; I admit I haven&#8217;t fully followed the ecosystem in recent years, but it&#8217;s always interesting to see initiatives of this type emerge.</p><ul><li><p><a href="https://encore.dev/blog/typescript-parser-wasm">Compiling a Typescript parser to WASM</a></p></li></ul><p>That&#8217;s it for this Weekly Tech. As usual, feel free to leave a comment / like / share if you feel like it. For my part, I&#8217;ll see you next week for a new update.</p><p>Until then, have a great weekend and a wonderful week.</p><p>See you Friday.</p>]]></content:encoded></item><item><title><![CDATA[Veille de la semaine du 13 Juillet 2026]]></title><description><![CDATA[Apple et la guerre de l'IA, Palantir serait-il le r&#233;el danger dans nos soci&#233;t&#233;s ? PostgreSQL se met &#224; Rust, la Playstation 5 sur PC via l'open-source, une couche de m&#233;moire pour vos agents, pr&#234;t(e)s ?]]></description><link>https://guikingone.substack.com/p/veille-de-la-semaine-du-13-juillet</link><guid isPermaLink="false">https://guikingone.substack.com/p/veille-de-la-semaine-du-13-juillet</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 17 Jul 2026 14:16:05 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/207415847/0249ab5131d786b31e540e7f686e823e.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>B</strong>ien le bonjour &#224; tous et &#224; toutes, j&#8217;esp&#232;re que tout le monde va bien.</p><p>Nouvelle semaine, nouvelle veille et cette semaine a &#233;t&#233; relativement charg&#233;e en terme d&#8217;open source, pas mal de petits sujets qui sont arriv&#233;s. Pour rester digeste, j&#8217;ai exclu volontairement exclu certains sujets, je pense notamment &#224; certains mod&#232;les qui peuvent &#234;tre sortis c&#244;t&#233; IA ou certaines technologies / librairies qui ont fait du bruit dans la semaine, je reviendrai dessus plus tard (probablement la semaine prochaine, ma todo list commence &#224; devenir effrayante).</p><p>Bref, assez parl&#233;, on se lance ?</p><div><hr></div><h1>Culture g&#233;n&#233;rale</h1><p><strong>O</strong>n commence donc avec la culture g&#233;n&#233;rale et notamment deux vid&#233;os qui m&#8217;ont particuli&#232;rement marqu&#233;, la premi&#232;re, tr&#232;s courte, est une vid&#233;o de MKBHD sur la question de savoir si Apple a perdu la course &#224; l&#8217;IA, je suis assez d&#8217;accord avec les conclusions de la vid&#233;o, je pense qu&#8217;Apple n&#8217;a pas perdu la course &#224; l&#8217;IA, ils ont peut-&#234;tre perdu la course au software, mais pas au hardware. Si l&#8217;on regarde le tout en d&#233;tail, beaucoup de hardware d&#8217;Apple permet de faire tourner l&#8217;IA, je pense notamment sur Mac ou autres (sans parler des iPhone&#8217;s ou m&#234;me de l&#8217;iPad) o&#249; l&#8217;on a des puces de plus en plus puissantes qui peuvent faire tourner des mod&#232;les complets (et m&#234;me de gros mod&#232;les sur les MBP Pro).</p><p>Je pense qu&#8217;Apple a pris une strat&#233;gie diff&#233;rente, &#224; savoir se concentrer sur le hardware et peu sur le software tout en int&#233;grant le tout dans le software via macOS / iOS (les derni&#232;res versions vont dans ce sens d&#8217;ailleurs) afin d&#8217;optimiser l&#8217;int&#233;gration au sein de son &#233;cosyst&#232;me, cela r&#233;duit la R&amp;D et permet de rester sur le segment hardware qu&#8217;Apple maitrise d&#233;j&#224; (notamment avec la sortie des puces M).</p><p>Autre sujet li&#233; de fa&#231;on &#8220;indirecte&#8221; avec un &#233;change avec le CEO de Palantir (je sais, la soci&#233;t&#233; pose probl&#232;me &#224; certaines personnes), pour le coup je me concentre assez peu sur la soci&#233;t&#233; et plut&#244;t sur ce qui a &#233;t&#233; dit dans l&#8217;entretien (je pense que l&#8217;on a tendance tr&#232;s tr&#232;s vite &#224; chercher des boucs &#233;missaires sans &#233;couter les arguments de chacuns), ce dernier a d&#8217;ailleurs l&#8217;avantage de mettre la lumi&#232;re sur le fait que parfois, nous pouvons &#234;tre malhonn&#234;tes avec ce que l&#8217;on aimerait avoir, nous voulons de la s&#233;curit&#233; mais pas les inconv&#233;nients, nous voulons des services plus engageants mais pas l&#8217;impact qu&#8217;ils auront sur nos vies ou sur les donn&#233;es qu&#8217;ils requi&#232;rent, plein de petits sujets comme &#231;a qui prennent sens, la vid&#233;o dure une heure, &#231;a s&#8217;&#233;coute relativement bien (m&#234;me si l&#8217;interviewer a tendance a tr&#232;s vite couper la parole et &#224; &#234;tre biais&#233; - &#224; mon sens -) et cela permet de mieux saisir la pens&#233;e qui dirige Palantir (et une partie de nos vies indirectement).</p><ul><li><p><a href="https://www.youtube.com/watch?v=PHpsdIHpLUE">So, what&#8217;s the deal with aliens?</a> - Youtube</p></li><li><p><a href="https://www.youtube.com/watch?v=eWKY0OnPByg">Did Apple lost the AI &#8220;race&#8221;? Maybe not</a> - Youtube</p></li><li><p><a href="https://www.youtube.com/watch?v=WAmNsgZHDzU">Solving the &#8220;public&#8221; health issue</a> - Youtube</p></li><li><p><a href="https://www.youtube.com/watch?v=xm0rGvvxVOM">Is Palantir a problem for democracies?</a> - Youtube</p></li></ul><div><hr></div><h1>Database</h1><p><strong>T</strong>out petit d&#233;tour du c&#244;t&#233; de la base de l&#8217;ann&#233;e pour d&#233;couvrir PgRust, une r&#233;&#233;criture de PostgreSQL en Rust, cette r&#233;&#233;criture passe l&#8217;int&#233;gralit&#233; des tests de PostgreSQL (ce qui est top et permet de s&#8217;assurer que rien n&#8217;est laiss&#233; sur le bas-c&#244;t&#233;) et est &#233;galement compatible avec WASM, une tr&#232;s bonne chose si vous souhaitez g&#233;rer votre base de donn&#233;es via le navigateur et optimiser vos appels r&#233;seau.</p><ul><li><p><a href="https://github.com/malisper/pgrust">PgRust, a rewrite of PostgreSQL in Rust</a> - Github</p></li></ul><div><hr></div><h1>Cloud computing / Linux</h1><p><strong>P</strong>etit d&#233;tour du c&#244;t&#233; du cloud / Linux, pour d&#233;couvrir B0nker, une alternative &#224; Docker, je n&#8217;ai pas eu l&#8217;occasion de l&#8217;essayer puisque je suis toujours en cours de test sur Dory, je vous laisse donc vous forger un avis.</p><p>Autre sujet qui aurait pu &#234;tre finalement &#224; la culture g&#233;n&#233;rale mais qui finit ici, SharpEmu, un &#233;mulateur exp&#233;rimental pour PlayStation 5 sur PC. Vous avez probablement suivi les actualit&#233;s r&#233;centes, PlayStation fera dispara&#238;tre ses support physique prochainement (&#224; ce sujet, vous savez probablement que GTA VI ne sera pas support&#233; en disque - le sens du timing comme on dit -) la communaut&#233; PC s&#8217;est donc demand&#233; si un &#233;mulateur pouvait voir le jour afin de r&#233;soudre le probl&#232;me, force est de constater que Claude est en support pour automatiser les t&#226;ches (je reviendrais sur ce sujet d&#232;s la semaine prochaine) et r&#233;duire le temps requis pour obtenir une version viable.</p><ul><li><p><a href="https://github.com/owlpharoah/b0nkers">B0nker, a lightweight containerization engine</a> - Github</p></li><li><p><a href="https://github.com/sharpemu/sharpemu">SharpEmu, an experimental Playstation 5 emulator</a> - Github</p></li></ul><div><hr></div><h1>Machine learning / AI</h1><p><strong>O</strong>n revient au machine learning / l&#8217;IA (le sujet devient un standard aujourd&#8217;hui) avec pas mal de petits sujets cette semaine et notamment Nowhere, une couche de m&#233;moire qui permet de lier vos documents avec vos agents, c&#8217;est moins &#8220;technique&#8221; que ce que peut faire Kage ou des solutions similaires telles qu&#8217;Honcho (d&#8217;ailleurs, &#224; ce sujet, j&#8217;ai enfin pu faire des tests avec Kage de fa&#231;on avanc&#233;e et je le recommande tr&#232;s chaudement l&#8217;outil, l&#8217;essayer, c&#8217;est l&#8217;adopter).</p><p>Autre outil (qui n&#8217;est pas tout &#224; fait un outil) avec PromptMac, un <em>helper</em> d&#233;velopp&#233; par Kevin Dunglas (membre de la Core Team Symfony), l&#8217;id&#233;e &#233;tant de vous simplifier la t&#226;che quant &#224; la cr&#233;ation d&#8217;un environnement agentic sur macOS, le process se r&#233;sumant &#224; : Vous achetez un mac, vous lancez la commande et le tout se d&#233;brouille pour installer ce dont vous avez besoin.</p><p>Dernier sujet avec GrokBuild (qui est sorti hier soir au moment o&#249; j&#8217;&#233;cris ces lignes), la couche de harness made in SpaceX AI. C&#8217;est &#233;crit en Rust (faites du Rust) et on pourrait le comparer &#224; Codex voir Claude Code sur certains aspects, je n&#8217;ai pas eu l&#8217;occasion de tester en d&#233;tails le tout, je vous laisse donc vous forger votre propre avis.</p><ul><li><p><a href="https://github.com/Ontos-AI/knowhere">Knowhere, a memory layer between documents and agents</a> - Github</p></li><li><p><a href="https://app.umans.ai/landing">Umans, an inference platform for open models</a></p></li><li><p><a href="https://github.com/omnigent-ai/omnigent">Omnigent, an open-source AI agent harness</a> - Github</p></li><li><p><a href="https://github.com/dunglas/prompt-mac">PromptMac, a single command to bootstrap an agentic environment on macOS</a> - Github</p></li><li><p><a href="https://github.com/workweave/router">Workweave Router, a model router for agentic systems</a> - Github</p></li><li><p><a href="https://github.com/xai-org/grok-build">Grok Build, SpaceXAI&#8217;s coding agent harness and TUI</a> - Github</p></li></ul><div><hr></div><h1>PHP</h1><p><strong>O</strong>n continue avec PHP et notamment la d&#233;couverte de PHPStorm en version Light (vous savez d&#233;j&#224; que j&#8217;ai migr&#233; sur Zed, je n&#8217;utilise plus PHPStorm mais je reste &#224; l&#8217;&#233;coute de ce que Jetbrains propose, on ne sait jamais), que dire de cette version Light ? Je ne suis pas super convaincu, je l&#8217;ai install&#233; sur ma machine pour tester et force est de constater qu&#8217;il reste relativement lourd pour ce qu&#8217;il propose, le LSP derri&#232;re n&#8217;est pas hyper performant (surtout quand des alternatives existent, sont open-source et plus performantes). Bref, est-ce que je pense que PHPStorm dans une version Light serait utile ? Discutable, ce type d&#8217;outil a souvent tendance &#224; couper ce qui apporte de la plus-value sans retravailler les fondations de l&#8217;outil, ce qui m&#232;ne &#224; un hybride mal d&#233;coup&#233; qui ne fait rien &#8220;comme pr&#233;vu&#8221;.</p><p>Autre sujet avec Thrun, un syst&#232;me de queue asynchrone, construit sur les fondations de TrueAsync, je n&#8217;ai pas eu l&#8217;occasion de le tester encore mais les benchmarks semblent convaincants, &#224; tester d&#232;s que possible.</p><ul><li><p><a href="https://phpstorm.dev/light-mode">PHPStorm is working on a &#8220;light&#8221; version</a></p></li><li><p><a href="https://github.com/SHWorX/beacon">Beacon, a modern framework</a> - Github</p></li><li><p><a href="https://github.com/yangusik/thrun">Thrun, an async queue worker</a> - Github</p></li></ul><div><hr></div><h1>Rust / Go</h1><p><strong>O</strong>n finit avec Rust avec notamment un tr&#232;s bon article sur la fa&#231;on dont Rust fragmente la m&#233;moire (certaines notions peuvent s&#8217;appliquer &#224; d&#8217;autres languages d&#8217;ailleurs) et comment r&#233;soudre ce genre de situation sans perdre votre patience.</p><ul><li><p><a href="https://kerkour.com/rust-high-performance-memory-fragmentation-allocations">Understanding and eliminating memory fragmentation</a></p></li><li><p><a href="https://github.com/embassy-rs/embassy/">Embassy, a modern framework for embedded</a> - Github</p></li></ul><p>Voici donc pour la veille de cette semaine, comme d&#8217;habitude, n&#8217;h&#233;sitez pas &#224; laisser un commentaire / like / partager si le coeur vous en dit, de mon c&#244;t&#233;, je vous donne rendez-vous d&#232;s la semaine prochaine pour une nouvelle veille.</p><p>D&#8217;ici-l&#224;, passez un excellent weekend ainsi qu&#8217;une tr&#232;s bonne semaine.</p><p>A vendredi.</p>]]></content:encoded></item><item><title><![CDATA[Weekly Tech - 13th of July 2026]]></title><description><![CDATA[Apple and the AI war, is Palantir the real danger to our societies? PostgreSQL turns to Rust, PS5 on PC via open-source, and a memory layer for your agents... ready?]]></description><link>https://guikingone.substack.com/p/weekly-tech-13th-of-july-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/weekly-tech-13th-of-july-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 17 Jul 2026 14:16:05 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/207416468/1fc5cfc59048764a81b44a83d549c8c6.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>H</strong>ello everyone, I hope you&#8217;re all doing well.</p><p>New week, new Weekly Tech, and this week has been relatively busy in terms of open source, with quite a few small topics popping up. To keep it digestible, I&#8217;ve deliberately excluded certain subjects&#8212;specifically some AI models that were released or certain technologies/libraries that made noise this week; I&#8217;ll get back to those later (probably next week, as my todo list is starting to become frightening).</p><p>Anyway, enough talking, shall we dive in?</p><div><hr></div><h1>General Knowledge</h1><p><strong>W</strong>e start with general knowledge, specifically two videos that particularly struck me. The first, very short one, is a video by MKBHD on whether Apple has lost the AI race. I quite agree with the video&#8217;s conclusions: I don&#8217;t think Apple has lost the AI race; they may have lost the software race, but not the hardware race. If we look at it in detail, much of Apple&#8217;s hardware allows AI to run, particularly on Macs or others (not to mention iPhones or even the iPad), where we have increasingly powerful chips capable of running complete models (and even large models on the MBP Pro).</p><p>I believe Apple took a different strategy: focusing on hardware and less on software, while integrating everything into the software via macOS / iOS (the latest versions are moving in this direction, by the way) to optimize integration within its ecosystem. This reduces R&amp;D and allows them to stay in the hardware segment that Apple already masters (notably with the release of the M chips).</p><p>Another topic &#8220;indirectly&#8221; related is an exchange with the CEO of Palantir (I know, the company is problematic for some people). In this case, I&#8217;m focusing very little on the company and more on what was said in the interview (I think we tend to look for scapegoats very quickly without listening to everyone&#8217;s arguments). The latter has the advantage of highlighting that sometimes we can be dishonest about what we want: we want security but not the drawbacks; we want more engaging services but not the impact they have on our lives or the data they require. Plenty of small topics like that that make sense; the video lasts an hour, it&#8217;s relatively easy to listen to (even if the interviewer tends to interrupt quickly and be biased&#8212;in my opinion), and it allows for a better grasp of the thinking that drives Palantir (and a part of our lives indirectly).</p><ul><li><p><a href="https://www.youtube.com/watch?v=PHpsdIHpLUE">So, what&#8217;s the deal with aliens?</a> - Youtube</p></li><li><p><a href="https://www.youtube.com/watch?v=eWKY0OnPByg">Did Apple lost the AI &#8220;race&#8221;? Maybe not</a> - Youtube</p></li><li><p><a href="https://www.youtube.com/watch?v=WAmNsgZHDzU">Solving the &#8220;public&#8221; health issue</a> - Youtube</p></li><li><p><a href="https://www.youtube.com/watch?v=xm0rGvvxVOM">Is Palantir a problem for democracies?</a> - Youtube</p></li></ul><div><hr></div><h1>Database</h1><p><strong>A</strong> quick detour into the database side to discover PgRust, a rewrite of PostgreSQL in Rust. This rewrite passes all PostgreSQL tests (which is great and ensures nothing is left behind) and is also compatible with WASM&#8212;a very good thing if you want to manage your database via the browser and optimize your network calls.</p><ul><li><p><a href="https://github.com/malisper/pgrust">PgRust, a rewrite of PostgreSQL in Rust</a> - Github</p></li></ul><div><hr></div><h1>Cloud computing / Linux</h1><p><strong>A</strong> quick detour into cloud / Linux to discover B0nker, an alternative to Docker. I haven&#8217;t had the chance to try it since I&#8217;m still in the middle of testing Dory, so I&#8217;ll let you form your own opinion.</p><p>Another subject that could have ended up in general knowledge but lands here: SharpEmu, an experimental emulator for PlayStation 5 on PC. You&#8217;ve probably followed recent news&#8212;PlayStation will phase out physical media soon (on that note, you probably know that GTA VI won&#8217;t be supported on disc&#8212;perfect timing, as they say). The PC community wondered if an emulator could emerge to solve the problem, and it&#8217;s clear that Claude is being used to automate tasks (I&#8217;ll come back to this subject next week) and reduce the time required to obtain a viable version.</p><ul><li><p><a href="https://github.com/owlpharoah/b0nkers">B0nker, a lightweight containerization engine</a> - Github</p></li><li><p><a href="https://github.com/sharpemu/sharpemu">SharpEmu, an experimental Playstation 5 emulator</a> - Github</p></li></ul><div><hr></div><h1>Machine learning / AI</h1><p><strong>W</strong>e return to machine learning / AI (the subject is becoming a standard today) with quite a few small topics this week, notably Knowhere, a memory layer that allows you to link your documents with your agents. It&#8217;s less &#8220;technical&#8221; than what Kage or similar solutions like Honcho can do (by the way, on that subject, I was finally able to perform advanced tests with Kage and I highly recommend the tool&#8212;try it and you&#8217;ll love it).</p><p>Another tool (which isn&#8217;t exactly a tool) is PromptMac, a <em>helper</em> developed by Kevin Dunglas (member of the Symfony Core Team). The idea is to simplify the task of creating an agentic environment on macOS; the process boils down to: You buy a Mac, you run the command, and everything takes care of installing what you need.</p><p>Final topic: GrokBuild (which came out last night as I write these lines), the harness layer made by SpaceX AI. It&#8217;s written in Rust (do some Rust) and could be compared to Codex or Claude Code in certain aspects. I haven&#8217;t had the chance to test it in detail, so I&#8217;ll let you form your own opinion.</p><ul><li><p><a href="https://github.com/Ontos-AI/knowhere">Knowhere, a memory layer between documents and agents</a> - Github</p></li><li><p><a href="https://app.umans.ai/landing">Umans, an inference platform for open models</a></p></li><li><p><a href="https://github.com/omnigent-ai/omnigent">Omnigent, an open-source AI agent harness</a> - Github</p></li><li><p><a href="https://github.com/dunglas/prompt-mac">PromptMac, a single command to bootstrap an agentic environment on macOS</a> - Github</p></li><li><p><a href="https://github.com/workweave/router">Workweave Router, a model router for agentic systems</a> - Github</p></li><li><p><a href="https://github.com/xai-org/grok-build">Grok Build, SpaceXAI&#8217;s coding agent harness and TUI</a> - Github</p></li></ul><div><hr></div><h1>PHP</h1><p><strong>W</strong>e continue with PHP, notably the discovery of PHPStorm in a Light version (you already know I migrated to Zed, I no longer use PHPStorm but I stay tuned to what JetBrains offers, you never know). What can I say about this Light version? I&#8217;m not super convinced; I installed it on my machine to test, and it remains relatively heavy for what it offers. The underlying LSP isn&#8217;t very high-performing (especially when alternatives exist, are open-source, and are more performant). In short, do I think PHPStorm in a Light version would be useful? Debatable; this type of tool often tends to cut what adds value without reworking the foundations of the tool, leading to a poorly sliced hybrid that doesn&#8217;t do anything &#8220;as planned.&#8221;</p><p>Another topic: Thrun, an asynchronous queue system built on the foundations of TrueAsync. I haven&#8217;t had the chance to test it yet, but the benchmarks seem convincing&#8212;to be tested as soon as possible.</p><ul><li><p><a href="https://phpstorm.dev/light-mode">PHPStorm is working on a &#8220;light&#8221; version</a></p></li><li><p><a href="https://github.com/SHWorX/beacon">Beacon, a modern framework</a> - Github</p></li><li><p><a href="https://github.com/yangusik/thrun">Thrun, an async queue worker</a> - Github</p></li></ul><div><hr></div><h1>Rust / Go</h1><p><strong>W</strong>e finish with Rust, notably a very good article on how Rust fragments memory (some notions can apply to other languages as well) and how to resolve these kinds of situations without losing your patience.</p><ul><li><p><a href="https://kerkour.com/rust-high-performance-memory-fragmentation-allocations">Understanding and eliminating memory fragmentation</a></p></li><li><p><a href="https://github.com/embassy-rs/embassy/">Embassy, a modern framework for embedded</a> - Github</p></li></ul><p>That&#8217;s all for this Weekly Tech. As usual, feel free to leave a comment / like / share if you feel like it. For my part, I&#8217;ll see you next week for a new watch.</p><p>Until then, have a great weekend and a wonderful week.</p><p>See you Friday.</p>]]></content:encoded></item><item><title><![CDATA[Veille de la semaine du 6 Juillet 2026]]></title><description><![CDATA[La Chine prouve que l'open source fonctionne, Apple ouvre Safari au MCP, Anthropic joue les dealers, et une nouvelle couche m&#233;moire pour agents d&#233;barque directement dans Git. du Docker, pr&#234;t(e)s ?]]></description><link>https://guikingone.substack.com/p/veille-de-la-semaine-du-6-juillet</link><guid isPermaLink="false">https://guikingone.substack.com/p/veille-de-la-semaine-du-6-juillet</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 10 Jul 2026 10:16:09 GMT</pubDate><content:encoded><![CDATA[<p><strong>B</strong>ien le bonjour &#224; tous et &#224; toutes, j&#8217;esp&#232;re que tout le monde va bien.</p><p>Nouvelle semaine, nouvelle Veille de la semaine et cette semaine, on reprend le rythme avec les sujets classiques, j&#8217;ai enfin eu le temps d&#8217;explorer certains sujets (sans parler de certains sujets sortis entre hier et aujourd&#8217;hui), je sais, je sais, OpenAI a encore sorti un mod&#232;le, &#224; ce sujet, je met volontairement de c&#244;t&#233; ce type d&#8217;informations car la v&#233;locit&#233; du secteur est trop grande pour &#234;tre suivie correctement, si jamais ce type d&#8217;informations &#8220;bouscule&#8221; le secteur, j&#8217;en parlerais mais pour le moment, consid&#233;rez que ce n&#8217;est pas prioritaire dans la veille.</p><p>Bref, assez parl&#233;, on se lance ?</p><div><hr></div><h1>Culture g&#233;n&#233;rale</h1><p><strong>B</strong>ien, attaquons par la culture g&#233;n&#233;rale et notamment par un podcast en fran&#231;ais sur l&#8217;avenir de l&#8217;IA et son impact au sein des diff&#233;rentes branches, sans oublier la concurrence qui s&#8217;installe du c&#244;t&#233; de la Chine. Je suis assez d&#8217;accord avec ce qui est dit dans le podcast (malgr&#233; certains d&#233;saccords mais rien de fondamental), la Chine est en train de montrer au reste du monde que l&#8217;open source fonctionne (et que donc, une approche <em>communiste</em> du sujet, c&#8217;est-&#224;-dire une mise en commun ce qui a &#233;t&#233; cr&#233;&#233; tout en conservant une approche capitaliste - dans le sens o&#249; la concurrence peut se lancer, le gouvernement &#233;tant juste l&#224; pour encadrer - est une approche viable dans ce type d&#8217;&#233;cosyst&#232;me tr&#232;s v&#233;loce), restons honn&#234;te, nous avons de l&#8217;open source en France / Europe (notamment via Mistral) mais nous restons tr&#232;s loin de la v&#233;locit&#233; des chinois (et je ne parle pas des USA), je persiste &#224; penser que si l&#8217;on passait moins de temps &#224; r&#233;guler et que les brides &#233;taient plus l&#226;ches, l&#8217;Europe (et non l&#8217;U.E, je fais la distinction car elle a son importance) serait aux avant-postes sur l&#8217;IA (dois-je rappelle que nous avons un parc nucl&#233;aire en France ? Non, vous voyez o&#249; je veux en venir).</p><p>Autre sujet de la semaine avec Apple rend disponible un serveur MCP via Safari, l&#8217;id&#233;e &#233;tant que le serveur MCP va pouvoir contr&#244;ler le navigateur, pour m&#233;moire, Chrome le fait d&#233;j&#224; (et a mis en avant WebMCP dont j&#8217;ai parl&#233; il y a peu de temps) mais n&#8217;ouvre pas le contr&#244;le &#8220;complet&#8221; du navigateur, peut-&#234;tre doit-on voir dans ce serveur la volont&#233; d&#8217;Apple d&#8217;ouvrir son &#233;cosyst&#232;me &#224; l&#8217;IA via un protocole standardis&#233; (et qui sait, peut-&#234;tre qu&#8217;&#224; l&#8217;avenir, Apple aura des mod&#232;les on-device ayant un acc&#232;s complet aux applications via MCP ...).</p><p>Dernier sujet de la semaine avec une vid&#233;o sur le comportement d&#8217;Anthropic et sa tendance &#224; se comporter comme un &#8220;dealer de drogue&#8221;, l&#8217;id&#233;e &#233;tant de vous rendre d&#233;pendant et &#8220;accro&#8221; tout en distillant des fonctionnalit&#233;s et du &#8220;bonus&#8221; sans r&#233;ellement vous le dire, cela permet de retirer ces derni&#232;res (coucou Fable) sans pr&#233;venir et vous forcer &#224; rester dans l&#8217;&#233;cosyst&#232;me, un sombre comportement quand on y pense.</p><ul><li><p><a href="https://www.youtube.com/watch?v=Ytj8MTd5AOs">What if China was the next step for open-source?</a> - Youtube (French)</p></li><li><p><a href="https://github.com/limelight-lang">Limelight, a runtime toolkit for building business-level programming languages</a> - Github</p></li><li><p><a href="https://webkit.org/blog/18136/introducing-the-safari-mcp-server-for-web-developers/">Apple unveiled an MCP server for Safari</a></p></li><li><p><a href="https://www.youtube.com/watch?v=b6V5F-mjUy8">Did Ferrari make a mistake with the Luce? Depends on the end goal of it</a> - Youtube (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=dcU2idUSRAA">What&#8217;s the difference between Vinyl, CD and Spotify?</a> - Youtube</p></li><li><p><a href="https://devblogs.microsoft.com/typescript/announcing-typescript-7-0/">Typescript 7.0 is out and it&#8217;s bloody fast</a></p></li><li><p><a href="https://www.youtube.com/watch?v=2lfMtyXCNKg">Is Anthropic a drug dealer when it comes to handling plans?</a> - Youtube</p></li></ul><div><hr></div><h1>Cloud computing / Linux</h1><p><strong>T</strong>out petit d&#233;tour du c&#244;t&#233; du cloud pour d&#233;couvrir Dory, une alternative pour macOS &#224; Docker, attention avant de penser &#224; basculer sur Dory, certains registries ne sont pas support&#233;s, idem pour certains types de r&#233;seau et pour certaines commandes Docker, le projet est assez r&#233;cent donc les correctifs arriveront s&#251;rement assez vite mais basculer tout de suite n&#8217;est probablement pas une bonne id&#233;e.</p><ul><li><p><a href="https://github.com/Augani/dory">Dory, a macOS native Docker alternative</a> - Github</p></li><li><p><a href="https://slicervm.com/">SlicerVM, full VMs anywhere</a></p></li></ul><div><hr></div><h1>Machine learning / AI</h1><p>Bien, quoi de neuf du c&#244;t&#233; du Machine learning? La semaine a &#233;t&#233; relativement charg&#233;e (OpenAI s&#8217;&#233;tant d&#233;cid&#233; &#224; rendre disponible ces mod&#232;les concurren&#231;ant Fable / Mythos) et elle a commenc&#233; par la d&#233;couverte de Kage, une couche de m&#233;moire pour vos agents qui s&#8217;installe directement dans les repositories Git (et qui est donc versionn&#233;e et partageable) et qui est construire autour du standard OKF rendu publique par Google, si vous avez lu les veilles pr&#233;c&#233;dentes, vous avez probablement vu Honcho, une couche m&#233;moire n&#233;cessitant une stack Docker (&#224; minima pour pouvoir l&#8217;emporter <em>partout</em>) et ne pouvant donc pas &#234;tre partag&#233;e facilement ou pouvant &#233;voluer avec le projet, en effet, Honcho n&#233;cessite des mod&#232;les d&#233;di&#233;s pour &#8220;construire&#8221; la m&#233;moire progressivement, Kage a le m&#233;rite de construire un graph versionn&#233; (tant dans Git que dans le code) sans d&#233;pendre de services externes (et donc, en r&#233;duisant les co&#251;ts), pour &#234;tre franc, je suis &#224; ma premi&#232;re semaine de tests, les r&#233;sultats valent donc &#8220;ce qu&#8217;ils valent&#8221; mais j&#8217;ai constat&#233; une am&#233;lioration des suggestions de mod&#232;les qui patinaient avec Honcho (et qui &#233;taient bien plus lents de par les appels externes), bref, si vous avez un moment, jetez-y un oeil.</p><p>Autre sujet de la semaine avec une vid&#233;o de youtube sur l&#8217;&#233;tat actuel du march&#233; de l&#8217;IA et notamment la concurrence entre mod&#232;le open-weight et &#8220;frontiers&#8221;, certains pensent que les mod&#232;les open-weight finiront par totalement rattraper les structures type Anthropic / OpenAI, jusqu&#8217;&#224; r&#233;cemment (et notamment GLM), j&#8217;&#233;tais dubitatif, je dois avouer que les arguments avanc&#233;s et l&#8217;&#233;tat actuel du march&#233; semblent aller dans ce sens.</p><ul><li><p><a href="https://github.com/Gitlawb/zero">Zero, a coding agent alternative to Claude Code / Opencode</a> - Github</p></li><li><p><a href="https://www.youtube.com/watch?v=pPg7hSKN43Y">What&#8217;s the current state of models and harness?</a> - Youtube (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=YWg_YCQEZhI">The battle between Open weight and Frontier models</a> - Youtube</p></li><li><p><a href="https://www.youtube.com/watch?v=q4YT5QPFrgQ">Could AI lead to new discoveries in science?</a> - Youtube (French)</p></li><li><p><a href="https://github.com/cosmicstack-labs/mercury-agent">Mercury Agent, a &#8220;soul-driven&#8221; AI agent</a> - Github</p></li><li><p><a href="https://github.com/kage-core/Kage">Kage, a memory layer for agents built on top of Git</a> - Github</p></li><li><p><a href="https://github.com/harbor-framework/harbor">Harbor, a framework for evaluating agents and models</a> - Github</p></li><li><p><a href="https://ollama.com/blog/all-aboard-open-models">Ollama is raising 88 millions to bring open models everywhere</a></p></li></ul><div><hr></div><h1>PHP</h1><p><strong>T</strong>r&#232;s l&#233;ger d&#233;tour du c&#244;t&#233; de php PHP d&#233;couvrir Clonio (&#224; quelques lettres pr&#234;t, le ton changeait rapidement ...), un outil en ligne de commande qui vous permet d&#8217;anonymiser et/ou de transf&#233;rer des donn&#233;es entre diff&#233;rents environnements, un outil indispensable si vous avez d&#233;j&#224; eu &#224; vous confronter &#224; des soucis de RGPD ou autres, l&#8217;id&#233;e est de r&#233;soudre ce type de situation sans d&#233;pendre de Docker (m&#234;me si une image Docker est disponible) tout en automatisant les t&#226;ches p&#233;nibles (anonymiser une DB n&#8217;est jamais agr&#233;able, surtout quand elle devient cons&#233;quente et contient des centaines de relations).</p><ul><li><p><a href="https://github.com/clonio-dev/clonio-cli">Clonio, a CLI to anonymize and transfer data between environments</a> - Github</p></li></ul><div><hr></div><h1>Tools</h1><p><strong>D</strong>ernier sujet de la semaine avec TablePro, un outil GUI pour acc&#233;der et se connecter &#224; vos bases de donn&#233;es SQL, Redis etc. Si vous avez un IDE ou un &#233;diteur de texte qui n&#8217;a pas le support pour la connexion sur la base de donn&#233;es (coucou Zed), cela permet d&#8217;effectuer les op&#233;rations classiques sans devoir passer en CLI (m&#234;me si soyons francs, rien ne vaut le debug en prod avec la CLI ...), &#224; surveiller si vous avez besoin d&#8217;un outil d&#233;di&#233;.</p><ul><li><p><a href="https://github.com/TableProApp/TablePro">TablePro, a free and open source database client</a> - Github</p></li></ul><p>Voici donc pour la veille de cette semaine, comme d&#8217;habitude, n&#8217;h&#233;sitez pas &#224; laisser un commentaire / like / partager si le coeur vous en dit, de mon c&#244;t&#233;, je vous donne rendez-vous d&#232;s la semaine prochaine pour une nouvelle veille.</p><p>D&#8217;ici-l&#224;, passez un excellent weekend ainsi qu&#8217;une tr&#232;s bonne semaine.</p><p>A vendredi.</p>]]></content:encoded></item><item><title><![CDATA[Weekly tech - 6th of July 2026]]></title><description><![CDATA[China proves that open source works, Apple opens Safari to MCP, Anthropic plays the dealer, and a new memory layer for agents lands directly in Git. Some Docker too, ready?]]></description><link>https://guikingone.substack.com/p/weekly-tech-6th-of-july-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/weekly-tech-6th-of-july-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 10 Jul 2026 10:16:09 GMT</pubDate><content:encoded><![CDATA[<p><strong>W</strong>ell, hello to all, I hope everyone is doing well.</p><p>New week, new Weekly Tech, and this week we&#8217;re getting back into the rhythm with the classic topics. I finally had the time to explore certain subjects (not to mention some that dropped between yesterday and today). I know, I know, OpenAI released yet another model, on that note, I&#8217;m deliberately setting this type of information aside because the sector&#8217;s velocity is too great to be followed properly. If ever this type of information &#8220;shakes up&#8221; the sector, I&#8217;ll mention it, but for the moment, consider that it isn&#8217;t a priority in the watch.</p><p>Anyway, enough talk, shall we get started?</p><div><hr></div><h1>Culture g&#233;n&#233;rale</h1><p><strong>R</strong>ight, let&#8217;s start with general knowledge and notably with a podcast in French about the future of AI and its impact across the various branches, without forgetting the competition emerging on China&#8217;s side. I largely agree with what&#8217;s said in the podcast (despite a few disagreements, but nothing fundamental): China is showing the rest of the world that open source works (and therefore that a communist approach to the subject, that is, pooling what has been created while retaining a capitalist approach, in the sense that competition can launch, the government being there merely to provide a framework - is a viable approach in this type of highly fast-moving ecosystem). Let&#8217;s stay honest: we have open source in France / Europe (notably via Mistral), but we remain very far from the velocity of the Chinese (and I&#8217;m not even talking about the USA). I maintain that if we spent less time regulating and if the constraints were looser, Europe (and not the E.U, I make the distinction because it matters) would be at the forefront of AI (do I need to remind you that we have a nuclear fleet in France? No, you see where I&#8217;m going with this).</p><p>Another topic of the week with Apple making an MCP server available via Safari, the idea being that the MCP server will be able to control the browser. For the record, Chrome already does this (and has showcased WebMCP, which I talked about a little while ago) but doesn&#8217;t open up &#8220;full&#8221; control of the browser. Perhaps we should see in this server Apple&#8217;s desire to open its ecosystem to AI via a standardized protocol (and who knows, perhaps in the future Apple will have on-device models with full access to applications via MCP...).</p><p>Last topic of the week with a video on Anthropic&#8217;s behavior and its tendency to act like a &#8220;drug dealer,&#8221; the idea being to make you dependent and &#8220;hooked&#8221; while doling out features and &#8220;bonuses&#8221; without really telling you, which makes it possible to withdraw the latter (hi there, Fable) without warning and force you to stay within the ecosystem, a dark behavior when you think about it.</p><ul><li><p><a href="https://www.youtube.com/watch?v=Ytj8MTd5AOs">What if China was the next step for open-source?</a> - Youtube (French)</p></li><li><p><a href="https://github.com/limelight-lang">Limelight, a runtime toolkit for building business-level programming languages</a> - Github</p></li><li><p><a href="https://webkit.org/blog/18136/introducing-the-safari-mcp-server-for-web-developers/">Apple unveiled an MCP server for Safari</a></p></li><li><p><a href="https://www.youtube.com/watch?v=b6V5F-mjUy8">Did Ferrari make a mistake with the Luce? Depends on the end goal of it</a> - Youtube (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=dcU2idUSRAA">What&#8217;s the difference between Vinyl, CD and Spotify?</a> - Youtube</p></li><li><p><a href="https://devblogs.microsoft.com/typescript/announcing-typescript-7-0/">Typescript 7.0 is out and it&#8217;s bloody fast</a></p></li><li><p><a href="https://www.youtube.com/watch?v=2lfMtyXCNKg">Is Anthropic a drug dealer when it comes to handling plans?</a> - Youtube</p></li></ul><div><hr></div><h1>Cloud computing / Linux</h1><p><strong>A</strong> very small detour to the cloud side to discover Dory, a macOS alternative to Docker. Be careful before thinking about switching to Dory: some registries aren&#8217;t supported, likewise for certain types of network and for certain Docker commands. The project is fairly recent, so<br>fixes will surely arrive fairly quickly, but switching right away probably isn&#8217;t a good idea.</p><ul><li><p><a href="https://github.com/Augani/dory">Dory, a macOS native Docker alternative</a> - Github</p></li><li><p><a href="https://slicervm.com/">SlicerVM, full VMs anywhere</a></p></li></ul><div><hr></div><h1>Machine learning / AI</h1><p><strong>R</strong>ight, what&#8217;s new on the Machine learning side? The week was relatively busy (OpenAI having decided to make available its models competing with Fable / Mythos) and it began with the discovery of Kage, a memory layer for your agents that installs directly into Git repositories (and which is therefore versioned and shareable) and is built around the OKF standard made public by Google. If you&#8217;ve read the previous watches, you&#8217;ve probably seen Honcho, a memory layer requiring a Docker stack (at minimum in order to be able to take it everywhere) and therefore unable to be shared easily or to evolve with the project. Indeed, Honcho requires dedicated models to &#8220;build&#8221; the memory progressively. Kage has the merit of building a versioned graph (both in Git and in the code) without depending on external services (and therefore reducing costs). To be honest, I&#8217;m on my first week of tests, so the results are worth &#8220;what they&#8217;re worth,&#8221; but I&#8217;ve observed an improvement in model suggestions that were stalling with Honcho (and which were much slower due to the external calls). Anyway, if you have a moment, take a look at it.</p><p>Another topic of the week with a YouTube video on the current state of the AI market and notably the competition between open-weight and &#8220;frontier&#8221; models. Some think that open-weight models will eventually completely catch up with structures like Anthropic / OpenAI. Until recently (and notably GLM), I was doubtful; I must admit that the arguments put forward and the current state of the market seem to go in that direction.</p><ul><li><p><a href="https://github.com/Gitlawb/zero">Zero, a coding agent alternative to Claude Code / Opencode</a> - Github</p></li><li><p><a href="https://www.youtube.com/watch?v=pPg7hSKN43Y">What&#8217;s the current state of models and harness?</a> - Youtube (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=YWg_YCQEZhI">The battle between Open weight and Frontier models</a> - Youtube</p></li><li><p><a href="https://www.youtube.com/watch?v=q4YT5QPFrgQ">Could AI lead to new discoveries in science?</a> - Youtube (French)</p></li><li><p><a href="https://github.com/cosmicstack-labs/mercury-agent">Mercury Agent, a &#8220;soul-driven&#8221; AI agent</a> - Github</p></li><li><p><a href="https://github.com/kage-core/Kage">Kage, a memory layer for agents built on top of Git</a> - Github</p></li><li><p><a href="https://github.com/harbor-framework/harbor">Harbor, a framework for evaluating agents and models</a> - Github</p></li><li><p><a href="https://ollama.com/blog/all-aboard-open-models">Ollama is raising 88 millions to bring open models everywhere</a></p></li></ul><div><hr></div><h1>PHP</h1><p><strong>A</strong> very light detour to the PHP side to discover Clonio (a few letters off and the tone would have changed quickly...), a command-line tool that lets you anonymize and/or transfer data between different environments, an indispensable tool if you&#8217;ve ever had to deal with GDPR issues or the like. The idea is to solve this type of situation without depending on Docker (even though a Docker image is available) while automating the tedious tasks (anonymizing a DB is never pleasant, especially when it becomes sizeable and contains hundreds of relations).</p><ul><li><p><a href="https://github.com/clonio-dev/clonio-cli">Clonio, a CLI to anonymize and transfer data between environments</a> - Github</p></li></ul><div><hr></div><h1>Tools</h1><p><strong>L</strong>ast topic of the week with TablePro, a GUI tool to access and connect to your SQL, Redis, etc. databases. If you have an IDE or a text editor that doesn&#8217;t have support for database connections (hi there, Zed), this lets you perform the classic operations without having to drop into the CLI (even though, let&#8217;s be honest, nothing beats debugging in prod with the CLI...), one to keep an eye on if you need a dedicated tool.</p><ul><li><p><a href="https://github.com/TableProApp/TablePro">TablePro, a free and open source database client</a> - Github</p></li></ul><p>So that&#8217;s it for this Weekly Tech. As usual, don&#8217;t hesitate to leave a comment / like / share if the heart takes you, and on my end, I&#8217;ll see you next week for a new Weekly Tech.</p><p>Until then, have an excellent weekend as well as a very good week.</p><p>See you Friday.</p>]]></content:encoded></item><item><title><![CDATA[Weekly Tech - 29th of June 2026]]></title><description><![CDATA[AI is likely still in its infancy, is the future more worrying? Fable 5 is coming back, but at what cost? PHP is turning to Rust for its tooling, for better or for worse? Ready?]]></description><link>https://guikingone.substack.com/p/weekly-tech-29th-of-june-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/weekly-tech-29th-of-june-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 03 Jul 2026 10:15:04 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/204803095/5229566ba65d540afb19a6cf1589f7ce.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>H</strong>ello everyone, I hope you are all doing well.</p><p>New week, new Weekly Tech, and this week I&#8217;ve decided to narrow the scope of the Weekly Tech because, to be honest, I didn&#8217;t have any major topics to cover (I mostly tested tools, and building a reasoned feedback takes time); some topics will therefore slide into the coming weeks depending on their priority.</p><p>Anyway, enough talking, shall we dive in?</p><div><hr></div><h1>General Knowledge</h1><p><strong>W</strong>e start with general knowledge, specifically a YouTube video discussing the fact that AI might not be as big of a problem as we think. Yes, criticisms may emerge on this subject, but fundamentally, wouldn&#8217;t the replication of the mechanism behind learning and intelligence be a far more significant issue? I don&#8217;t know where this research is heading or, above all, what it will enable us to produce, but the video reminded me of a presentation by John Carmack - one of the fathers of Doom - (I can&#8217;t recall the exact video, it should be on YouTube) who explained that to develop a neural network, using video games and achieving success within them might be one of the keys to going beyond the current state of the art (notably because the reward is stronger and more immediate than during standard training). Definitely worth digging into.</p><p>Another more general topic on economics, specifically a debate with Larry Lepard (a former &#8220;gold bug,&#8221; as they say in the field) who questions a return to &#8220;sound money.&#8221; The subject fascinates me due to its impacts and what it implies for the current state of the economy (as many libertarians say, the root of your problem isn&#8217;t capitalism but the currency: &#8220;change the money, change the world&#8221;). Anyway, I&#8217;ll let you take a look to form your own opinion.</p><ul><li><p><a href="https://www.youtube.com/watch?v=yGjy8La0kdU">What if AI wasn&#8217;t the real deal?</a> - Youtube (French)</p></li><li><p><code>l0g.fr</code><a href="https://l0g.fr/">, an online mapping system for economic risks</a></p></li><li><p><a href="https://www.youtube.com/watch?v=QjbrzelVrTE">The return to sound money will be hard</a> - Youtube (French)</p></li></ul><div><hr></div><h1>Machine Learning / AI</h1><p><strong>W</strong>hat&#8217;s new on the Machine Learning / AI side? For this week, you&#8217;ve probably seen it (the information was officialized last night as I write this), Fable 5 is back. I won&#8217;t dwell on Fable 5 because I think there is a general mockery coming from Anthropic on this matter, but let&#8217;s remain polite; the model returned heavily throttled (to say it&#8217;s completely blocked on most requests), and some topics will likely surface in the coming weeks, so keep an eye on it.</p><p>Aside from that, we can note the release of Lumo 2.0, which is still not open source in the strict sense (the application remains open-source but not the model, which is curious when you consider they are likely using an open-weight model like Ministral or Qwen) but adds image support (where v1 only supported text) and better tool management.</p><p>Last topic with SocratiCode, an MCP server that allows you to plug vector search and models into your codebase. Very interesting if you want to map your code, the different layers, the various components of your projects, and so on. A small but important detail: it all works &#8220;by default&#8221; via Ollama (meaning zero risk of data leaks as long as you use a local model) and consumes almost no resources to run (couple this with a memory layer like Honcho and it becomes a whole different story...).</p><ul><li><p><a href="https://github.com/safishamsi/graphify">Graphify, use your AI assistant to map your entire project into a knowledge graph</a> - Github</p></li><li><p><a href="https://x.com/asklumo/status/2071949404836122850">Lumo 2.0 is out</a> - X</p></li><li><p><a href="https://www.anthropic.com/news/redeploying-fable-5">Fable 5 is back... partially</a></p></li><li><p><a href="https://github.com/usestrix/strix">Strix, AI agent for penetration testing</a> - Github</p></li><li><p><a href="https://github.com/edgee-ai/edgee">Edgee, an open-source LLM gateway</a> - Github</p></li><li><p><a href="https://github.com/ghostwright/phantom">Phantom, an AI co-worker</a> - Github</p></li><li><p><a href="https://github.com/giancarloerra/SocratiCode">SocratiCode, an MCP server for your codebase</a> - Github</p></li></ul><div><hr></div><h1>PHP</h1><p><strong>L</strong>ast topic with PHP, specifically a subject discovered on X regarding the rewrite&#8212;or at least the migration&#8212;of Psalm from PHP to Pzoom, its Rust equivalent. It&#8217;s always interesting to see a migration to Rust, especially when talking about tools like PHPStan; these tend to prioritize PHP because of its parser (available on Github) and set aside performance as well as library evolution. Furthermore, this comes &#8220;at the right time&#8221; recalling recent events regarding generics support in PHP and the words from last year within the PHP Foundation about using Rust to extend the language.</p><p>PHP is not dead and will not die tomorrow. Can Rust push it forward into 2026 with a stable, performant, and resilient ecosystem? Only time will tell.</p><ul><li><p><a href="https://mattbrown.dev/articles/from-psalm-to-pzoom">From Psalm to Pzoom</a></p></li><li><p><a href="https://github.com/muglug/pzoom">Pzoom, a rewrite of PHPStan in Rust</a> - Github</p></li><li><p><a href="https://github.com/mnapoli/vibephp">VibePHP, a PHP engine that runs on vibes, not code</a> - Github</p></li></ul><p>That&#8217;s it for this week&#8217;s watch. As usual, feel free to leave a comment / like / share if you feel like it. For my part, I&#8217;ll see you next week for a new update.</p><p>Until then, have a great weekend and a wonderful week.</p><p>See you Friday.</p>]]></content:encoded></item><item><title><![CDATA[Veille de la semaine du 29 Juin 2026]]></title><description><![CDATA[L'IA n'est probablement qu'&#224; ses balbutiements, l'avenir serait-il plus inqui&#233;tant ? Fable 5 revient mais &#224; quel prix ? PHP se met &#224; Rust pour ses outils, pour le meilleur et le pire ? Pr&#234;t(e)s ?]]></description><link>https://guikingone.substack.com/p/veille-de-la-semaine-du-29-juin-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/veille-de-la-semaine-du-29-juin-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 03 Jul 2026 10:15:04 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/204802566/a4cb520cd943388c12abb40cafe033b8.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>B</strong>ien le bonjour &#224; tous et &#224; toutes, j&#8217;esp&#232;re que tout le monde va bien.</p><p>Nouvelle semaine, nouvelle Veille de la semaine et cette semaine, j&#8217;ai d&#233;cid&#233; de r&#233;duire le p&#233;rim&#232;tre de la veille car autant &#234;tre franc, je n&#8217;ai pas eu de gros sujets &#224; voir cette semaine (j&#8217;ai majoritairement test&#233; des outils et construire un retour argument&#233; prend du temps), certains sujets vont donc glisser sur les semaines &#224; venir en fonction de leur priorit&#233;.</p><p>Bref, assez parl&#233;, on se lance ?</p><div><hr></div><h1>Culture g&#233;n&#233;rale</h1><p><strong>O</strong>n commence par la culture g&#233;n&#233;rale et notamment une vid&#233;o Youtube discutant du fait que l&#8217;IA n&#8217;est peut-&#234;tre pas si gros probl&#232;me que ce que l&#8217;on pense, oui, des critiques peuvent &#233;merger &#224; ce sujet mais au fond, est-ce que la r&#233;plication du m&#233;canisme derri&#232;re l&#8217;apprentissage et l&#8217;intelligence ne serait pas un probl&#232;me bien plus important ? Je ne sais pas o&#249; ira cette recherche et surtout, ce qu&#8217;elle permettra de produire mais la vid&#233;o m&#8217;a rappel&#233; une pr&#233;sentation de John Carmack - un des papa des Doom - (je n&#8217;ai plus la vid&#233;o en t&#234;te, elle doit se retrouver sur Youtube) qui expliquait que pour d&#233;velopper un r&#233;seau neural, l&#8217;utilisation des jeux vid&#233;os et la r&#233;ussite au sein de ces derniers &#233;tait peut-&#234;tre une des clefs pour aller plus loin que l&#8217;&#233;tat actuel de l&#8217;art (notamment car la r&#233;compense est plus forte et plus imm&#233;diate que durant un entrainement standard), &#224; creuser.</p><p>Autre sujet plus g&#233;n&#233;ral sur l&#8217;&#233;conomie et notamment un d&#233;bat avec Larry Lepard (un ancien &#8220;gold bug&#8221; comme on dit dans le milieu) qui se pose la question d&#8217;un retour &#224; une &#8220;sound money&#8221; (ou monnaie saine en bon fran&#231;ais), le sujet me passionne de par ses impacts et ce qu&#8217;il induit pour l&#8217;&#233;tat actuel de l&#8217;&#233;conomie (comme disent beaucoup de lib&#233;raux, le fond de votre probl&#232;me n&#8217;est pas le capitalisme mais la monnaie, &#8220;change the money, change world&#8221;), bref, je vous laisse y jeter un oeil pour vous forgez votre propre avis.</p><ul><li><p><a href="https://www.youtube.com/watch?v=yGjy8La0kdU">What if AI wasn&#8217;t the real deal?</a> - Youtube (French)</p></li><li><p><code>l0g.fr</code><a href="https://l0g.fr/">, an online mapping system for economic risks</a></p></li><li><p><a href="https://www.youtube.com/watch?v=QjbrzelVrTE">The return to sound money will be hard</a> - Youtube (French)</p></li></ul><div><hr></div><h1>Machine learning / AI</h1><p><strong>Q</strong>uoi de neuf du c&#244;t&#233; du machine Learning / IA ? Pour cette semaine, vous l&#8217;avez probablement vu (l&#8217;information a &#233;t&#233; officialis&#233;e hier soir au moment o&#249; j&#8217;&#233;cris), Fable 5 est revenu, je ne m&#8217;&#233;talerai pas sur le sujet de Fable 5 car que je pense qu&#8217;il y a un <em>foutage</em> de gueule g&#233;n&#233;ral de la part d&#8217;Anthropic &#224; ce sujet mais restons courtois, le mod&#232;le est revenu en &#233;tant &#233;norm&#233;ment brid&#233; (pour ne pas dire compl&#232;tement bloqu&#233; sur la majorit&#233; des demandes), des sujets vont probablement faire surface dans les semaines &#224; venir, &#224; surveiller.</p><p>En dehors de &#231;a, on peut noter la sortie de Lumo 2.0 qui n&#8217;est toujours pas open source &#224; proprement parler (l&#8217;application reste open-source mais pas le mod&#232;le, ce qui est cocasse quand on r&#233;fl&#233;chis au fait qu&#8217;ils utilisent probablement un mod&#232;le open-weight type Ministral ou Qwen) mais qui ajoute le support des images (l&#224; o&#249; la v1 ne supportait que le texte) et une meilleure gestion des outils.</p><p>Dernier sujet avec SocratiCode, un serveur MCP qui permet de venir brancher de la recherche vectorielle et des mod&#232;les sur votre code base, tr&#232;s int&#233;ressant si jamais vous souhaitez cartographier votre code, les diff&#233;rentes couches, les diff&#233;rents composants de vos projets et ainsi de suite, petit d&#233;tail qui a son importance, le tout fonctionne &#8220;par d&#233;faut&#8221; via Ollama (donc z&#233;ro risques de fuite de donn&#233;es tant que vous utilisez un mod&#232;le local) et ne consomme quasiment rien en ressources pour tourner (couplez le tout avec une couche de m&#233;moire type Honcho et le sujet devient tout autre chose ...).</p><ul><li><p><a href="https://github.com/safishamsi/graphify">Graphify, use your AI assistant to map your entire project into a knowledge graph</a> - Github</p></li><li><p><a href="https://x.com/asklumo/status/2071949404836122850">Lumo 2.0 is out</a> - X</p></li><li><p><a href="https://www.anthropic.com/news/redeploying-fable-5">Fable 5 is back... partially</a></p></li><li><p><a href="https://github.com/usestrix/strix">Strix, AI agent for penetration testing</a> - Github</p></li><li><p><a href="https://github.com/edgee-ai/edgee">Edgee, an open-source LLM gateway</a> - Github</p></li><li><p><a href="https://github.com/ghostwright/phantom">Phantom, an AI co-worker</a> - Github</p></li><li><p><a href="https://github.com/giancarloerra/SocratiCode">SocratiCode, an MCP server for your codebase</a> - Github</p></li></ul><div><hr></div><h1>PHP</h1><p><strong>D</strong>ernier sujet avec PHP et notamment avec un sujet d&#233;couvert sur X qui est la r&#233;&#233;criture, tout du moins la migration de Psalm en version PHP vers Pzoom, son &#233;quivalent en Rust, toujours int&#233;ressant de voir une migration se faire vers Rust notamment quand on parle d&#8217;outils type PHPStan, ces derniers ont tendance &#224; privil&#233;gier PHP de par son parser (disponible sur Github) et &#224; mettre de c&#244;t&#233; les performances ainsi que l&#8217;&#233;volution de la librairie. En outre, le sujet arrive &#8220;au bon moment&#8221; quand on se souvient les &#233;v&#232;nements r&#233;cents sur le support des g&#233;n&#233;riques en PHP et les mots de l&#8217;ann&#233;e derni&#232;re au sein de la Fondation PHP sur l&#8217;usage de Rust pour &#233;tendre le language.</p><p>PHP n&#8217;est pas mort et ne mourra pas demain, Rust peut-il le faire passer en 2026 avec un &#233;cosyst&#232;me stable, performante et r&#233;silient ? Seul l&#8217;avenir nous le dira.</p><ul><li><p><a href="https://mattbrown.dev/articles/from-psalm-to-pzoom">From Psalm to Pzoom</a></p></li><li><p><a href="https://github.com/muglug/pzoom">Pzoom, a rewrite of PHPStan in Rust</a> - Github</p></li><li><p><a href="https://github.com/mnapoli/vibephp">VibePHP, a PHP engine that runs on vibes, not code</a> - Github</p></li></ul><p>Voici donc pour la veille de cette semaine, comme d&#8217;habitude, n&#8217;h&#233;sitez pas &#224; laisser un commentaire / like / partager si le coeur vous en dit, de mon c&#244;t&#233;, je vous donne rendez-vous d&#232;s la semaine prochaine pour une nouvelle veille.</p><p>D&#8217;ici-l&#224;, passez un excellent weekend ainsi qu&#8217;une tr&#232;s bonne semaine.</p><p>A vendredi.</p>]]></content:encoded></item><item><title><![CDATA[Weekly Tech - 22nd of June 2026]]></title><description><![CDATA[AI at home or decentralized, PHP's place in the AI ecosystem, and not to be forgotten: controlling your fans under Windows. This week's edition is a bit lighter but still resolutely technical... ready]]></description><link>https://guikingone.substack.com/p/weekly-tech-22nd-of-june-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/weekly-tech-22nd-of-june-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 26 Jun 2026 10:16:02 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/203587097/7e7bc820f81ede5ce65d06c1826fac48.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>G</strong>ood morning everyone, I hope you are all doing well.</p><p>New week, new Weekly Tech. To be honest, there are few topics this week; I&#8217;ve tried to keep it relatively short so that it remains readable and pleasant for everyone.</p><p>This week is mostly about machine learning and PHP (as these were the two big subjects that occupied me this week). Some topics will return next week (to be perfectly honest, I have a gigantic waiting list).</p><p>Anyway, I&#8217;ll stop talking and let&#8217;s dive in.</p><div><hr></div><h1>General Knowledge</h1><p><strong>W</strong>e start with general knowledge this week, specifically with two topics that could have ended up in machine learning but that I&#8217;m placing here because they fall (in my opinion) under general knowledge.</p><p>First topic: building your own agent via an infrastructure under your control to reduce costs. Now, let&#8217;s be honest, we aren&#8217;t talking about massive models (like Kimi or GLM), but models substantial enough to develop properly and have what&#8217;s needed for daily tasks. The tutorial also explains how to &#8220;publish&#8221; the agent and make it accessible outside your infrastructure - a must when you&#8217;re traveling and want to maintain control over your data.</p><p>Another topic of the week is Unreal Engine, which will open its engine (in v5.8) via an MCP server (making it available for Claude and the like) to simplify the creation of scenes/nodes. (If you&#8217;ve ever played with UE documentation, you probably know it&#8217;s... equivalent to Dante&#8217;s Inferno). The arrival of models can only be a benefit in breaking down capability barriers (not everyone masters C/C++) and allowing anyone to develop prototypes quickly.</p><ul><li><p><a href="https://minimachine.ulrichrozier.com/">Building your own local AI agent, the proper way</a></p></li><li><p><a href="/__u/preuvesdetravail.substack.com/p/limite-21-millions-but-originel-bitcoin">The 21 millions cap of Bitcoin wasn&#8217;t there at its birth</a> (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=AlV__BFg8qk">Unreal Engine is opening its core to MCP</a> - Youtube</p></li></ul><div><hr></div><h1>Machine Learning / AI</h1><p><strong>W</strong>e take a detour into machine learning for two small topics (or three, let&#8217;s be generous). The first is the use of Claude Code to build an action plan. Yes, I know, I know&#8212;Claude already has a <code>/plan</code> mode that allows the model to &#8220;think&#8221; about the initial request and propose a measured implementation. The idea here is to build a documented plan with back-and-forths between you and the model to refine the request and propose a final implementation. This prevents the model from having to &#8220;think/reason&#8221; through the final step, thereby avoiding unnecessary iterations and potentially reducing token consumption (coupling this with a memory system makes consumption drop drastically).</p><p>Another topic is an article explaining how to reduce hallucinations in models. Some techniques are debatable (I still believe RAG has its place in current stacks, despite what market giants say), and I don&#8217;t think that widening the context window further will solve the problems (you can put a V6 engine in a Twingo, but that doesn&#8217;t make it a Porsche). Since I am currently testing some solutions (notably Honcho) in Claude Code and in more complex pipelines, I will probably have feedback soon.</p><p>I&#8217;ll finish with a very small topic (which I touched upon in a YouTube video a few weeks ago) called Gonka, a decentralized architecture for models. An alternative created by Jack, the founder of Twitter, already existed (which I mentioned in a previous watch). Gonka is developed by former Instagram engineers and aims to be decentralized &#8220;Bitcoin-style.&#8221; I know decentralization can be a scary topic, but with recent developments (hello Fable, gone from the front...), it seems the subject has become more than important - even indispensable - if you want to maintain a minimum of control over your models and their resilience to censorship.</p><ul><li><p><a href="https://eurollm.io/">EuroLLM, LLMs made in Europe</a></p></li><li><p><a href="https://github.com/nvk/llm-wiki">LLM-wiki, a LLM-compiled knowledge base for any agent</a> - Github</p></li><li><p><a href="https://github.com/gonka-ai/gonka/">Gonka, a decentralized AI architecture</a> - Github</p></li><li><p><a href="https://boristane.com/blog/how-i-use-claude-code/">Using Claude Code by building a detailed plan</a></p></li><li><p><a href="https://dev.to/aws/stop-ai-agent-hallucinations-4-essential-techniques-2i94">Stop agents hallucinations</a> - dev.to</p></li><li><p><a href="https://github.com/VectifyAI/PageIndex">PageIndex, a vectorless, reasoning-based RAG</a> - Github</p></li></ul><div><hr></div><h1>PHP</h1><p><strong>W</strong>e go back to the classics with PHP, specifically an article on the current AI ecosystem (in the broad sense) within PHP. There are many existing libraries, some I&#8217;ve just discovered and others I already knew (notably TransformersPhp - a must if you want local models directly in your Symfony/Laravel applications). The article explains very well that PHP is already ready for AI but continues to suffer from the usual clich&#233;s - a &#8220;real shame&#8221; considering PHP&#8217;s deployment statistics worldwide.</p><p>Moving on to PicklePantherBundle, a Symfony bundle that allows implementing the &#8220;Behat approach&#8221; within Panther (in practice, the syntax strongly reminds me of the E2E testing tool developed by Blackfire a few years ago). I admit I&#8217;m not a fan of YAML syntax (simply because I don&#8217;t like the syntax; nothing personal against the bundle), but seeing extensions for Panther is always a pleasure and helps keep the library alive (even if Playwright has shaken up the ecosystem quite a bit).</p><p>Last small topic on optimization (I read the article a week ago and could have shared it then, but the previous watch was full enough) concerning PHP-FPM, specifically its scaling. This is indispensable if you regularly see your infrastructure crash after thousands of requests.</p><ul><li><p><a href="https://linuxblog.io/php-fpm-tuning-using-pm-static-max-performance/">Tuning PHP-FPM for maximum performances</a></p></li><li><p><a href="https://github.com/Amoifr/pickle-panther-bundle">PicklePantherBundle, a YAML-driven E2E testing engine on top of Panther</a> - Github</p></li><li><p><a href="https://zadroweb.com/blog/php-ai-agents/">PHP is already built for the AI era</a></p></li><li><p><a href="https://github.com/Kocal/BiomeJsBundle">A Symfony Bundle to use Biome.js as a frontend linter</a> - Github</p></li><li><p><a href="https://blog.phalcon.io/post/phalcon-v6-0-0alpha1-released">Phalcon is now a &#8220;full PHP&#8221; framework</a></p></li><li><p><a href="https://www.exakat.io/php-and-ai-what-actually-exists-at-the-language-level/">The current state of AI at language level</a></p></li><li><p><a href="https://github.com/symfony/webpack-encore/releases/tag/v7.0.0">Webpack Encore 7 is out</a> - Github</p></li><li><p><a href="https://docs.volt-test.com/blog/php-stress-testing-tool/">Stress testing PHP applications using Volt Test</a></p></li><li><p><a href="https://github.com/php-opcua">PHP OPC UA, a pure implementation of the OPC UA protocol stack</a> - Github</p></li></ul><div><hr></div><h1>Tools</h1><p><strong>W</strong>e wrap up this watch with tools, specifically FanControl, a utility I discovered by chance on X a week ago (if I&#8217;m not mistaken) that allows you to bypass &#8220;closed source&#8221; (and by extension, proprietary) solutions for controlling your fans (GPU included). I am currently testing it, so I&#8217;ll probably have an opinion soon. If you have the time, it&#8217;s worth testing (and it&#8217;s free).</p><ul><li><p><a href="https://github.com/tobi/qmd">qmd, a mini CLI search engine for your documents and others</a> - Github</p></li><li><p><a href="https://github.com/Rem0o/FanControl.Releases">FanControl, a fan controller for your computer (Windows only)</a> - Github</p></li><li><p><a href="https://github.com/originalankur/maptoposter">MapToPoster, transform your favorite cities onto minimalist designs</a> - Github</p></li></ul><p>That&#8217;s all for this Weekly Tech. As usual, feel free to leave a comment, like, or share if you feel like it. For my part, I&#8217;ll see you next week for a new update.</p><p>Until then, have an excellent weekend and a great week.</p><p>See you Friday.</p>]]></content:encoded></item><item><title><![CDATA[Veille de la semaine du 22 Juin 2026]]></title><description><![CDATA[L'IA &#224; la maison ou de fa&#231;on d&#233;centralis&#233;e, PHP au sein de l'&#233;cosyst&#232;me IA sans oublier le contr&#244;le de vos ventilateurs sous Windows, cette semaine se veut l&#233;g&#232;re mais r&#233;solument technique, pr&#234;t(e)s ?]]></description><link>https://guikingone.substack.com/p/veille-de-la-semaine-du-22-juin-2026</link><guid isPermaLink="false">https://guikingone.substack.com/p/veille-de-la-semaine-du-22-juin-2026</guid><dc:creator><![CDATA[Guillaume Loulier]]></dc:creator><pubDate>Fri, 26 Jun 2026 10:16:01 GMT</pubDate><enclosure url="https://api.substack.com/feed/podcast/203586451/9b73a014ab07896ed1a117c6494e0f84.mp3" length="0" type="audio/mpeg"/><content:encoded><![CDATA[<p><strong>B</strong>ien le bonjour &#224; tous et &#224; toutes, j&#8217;esp&#232;re que tout le monde va bien.</p><p>Nouvelle semaine, nouvelle Veille de la semaine, pour &#234;tre franc, peu de sujets cette semaine, j&#8217;ai essay&#233; de faire relativement court de sorte que ce soit lisible et agr&#233;able pour tout le monde.</p><p>Cette semaine, majoritairement du machine learning et du PHP (car ce sont les deux gros sujets qui m&#8217;ont occup&#233;s cette semaine), certains sujets reviendront la semaine prochaine (pour &#234;tre tout &#224; fait honn&#234;te, j&#8217;ai une liste d&#8217;attente gigantesque).</p><p>Bref, je m&#8217;arr&#234;te de parler et on se lance.</p><div><hr></div><h1>Culture g&#233;n&#233;rale</h1><p><strong>O</strong>n commence par la culture g&#233;n&#233;rale cette semaine et notamment par deux sujets qui auraient pu finir dans le machine learning mais que je glisse ici parce qu&#8217;ils rel&#232;vent (&#224; mon sens) de la culture g&#233;n&#233;rale.</p><p>Premier sujet avec la construction de son propre agent via une infrastructure sous votre contr&#244;le permettant de r&#233;duire les co&#251;ts, alors bien entendu, soyons honn&#234;tes, nous ne parlons de tr&#232;s gros mod&#232;les (tels que Kimi ou GLM) mais des mod&#232;les suffisamment cons&#233;quents pour pouvoir d&#233;velopper correctement et avoir de ce qu&#8217;il faut pour les t&#226;ches du quotidien, le tutoriel explique &#233;galement comment &#8220;publier&#8221; l&#8217;agent et le rendre accessible en dehors de votre infrastructure, un must quand vous &#234;tes en d&#233;placement et souhaitez conserver le contr&#244;le sur vos donn&#233;es.</p><p>Autre sujet de la semaine avec Unreal Engine qui va ouvrir son moteur (en v5.8) via un serveur MCP (donc disponible pour Claude et consorts) afin de simplifier la cr&#233;ation de sc&#232;ne / de noeuds (si vous avez d&#233;j&#224; jou&#233; avec la documentation UE, vous savez probablement qu&#8217;elle est ... &#233;quivalente &#224; l&#8217;enfer de Dante), voir arriver les mod&#232;les ne peut faire que du bien afin de faire sauter des barri&#232;res de capacit&#233; (tout le monde ne maitrise pas C / C++) et permettre &#224; tout un chacun de d&#233;velopper des prototypes rapidement.</p><ul><li><p><a href="https://minimachine.ulrichrozier.com/">Building your own local AI agent, the proper way</a></p></li><li><p><a href="/__u/preuvesdetravail.substack.com/p/limite-21-millions-but-originel-bitcoin">The 21 millions cap of Bitcoin wasn&#8217;t there at its birth</a> (French)</p></li><li><p><a href="https://www.youtube.com/watch?v=AlV__BFg8qk">Unreal Engine is opening its core to MCP</a> - Youtube</p></li></ul><div><hr></div><h1>Machine learning / AI</h1><p><strong>O</strong>n fait un d&#233;tour du c&#244;t&#233; du machine learning pour deux petits sujets (voir trois, soyons g&#233;n&#233;reux), le premier &#233;tant l&#8217;utilisation de Claude Code afin de construire un plan d&#8217;action, oui, je sais je sais, Claude a d&#233;j&#224; un mode <code>/plan</code> qui permet au mod&#232;le de &#8220;r&#233;fl&#233;chir&#8221; &#224; la demande initiale et de proposer une impl&#233;mentation mesur&#233;e, l&#8217;id&#233;e ici est de construire un plan document&#233; avec des allers-retours entre le mod&#232;le et vous afin d&#8217;affiner la demande et proposer une impl&#233;mentation finale qui &#233;vite au mod&#232;le de &#8220;penser / r&#233;fl&#233;chir&#8221; &#224; cette derni&#232;re, cela permet donc d&#8217;&#233;viter les allers-retours inutiles et peut r&#233;duire la consommation de jetons (coupler le tout avec un syst&#232;me de m&#233;moire et la consommation chute drastiquement).</p><p>Autre sujet avec un article expliquant comment r&#233;duire les hallucinations au niveau des mod&#232;les, certaines techniques qui sont discutables (je persiste &#224; penser que le RAG a son int&#233;r&#234;t dans les stacks actuels, malgr&#233; ce qu&#8217;en disent les g&#233;ants du march&#233;) et je pense que ce n&#8217;est pas en ouvrant de plus en plus la fen&#234;tre de contexte que les probl&#232;mes se r&#233;soudront (vous pouvez mettre un V6 dans une Twingo, &#231;a n&#8217;en fait pas une Porsche). Etant actuellement entrain de tester certaines solutions (notamment Honcho) dans Claude Code et dans des pipelines plus complexes, j&#8217;aurais probablement un retour d&#8217;exp&#233;rience &#224; faire prochainement.</p><p>Je termine avec un tout petit sujet (que j&#8217;avais abord&#233; via une vid&#233;o YouTube il y a quelques semaines) nomm&#233; Gonka, une architecture pour les mod&#232;les de fa&#231;on d&#233;centralis&#233;e, il existait d&#233;j&#224; une alternative cr&#233;&#233;e par Jack, le fondateur de Twitter (et dont j&#8217;avais parl&#233; dans la veille). Gonka est d&#233;velopp&#233; par des anciens ing&#233;nieurs d&#8217;Instagram et se veut d&#233;centralis&#233;e &#8220;&#224; la fa&#231;on de Bitcoin&#8221;, je sais que le sujet de la d&#233;centralise fait parfois peur mais avec les d&#233;veloppements r&#233;cents (coucou Fable, disparu au front ...), il semble que le sujet soit devenu plus qu&#8217;important voir indispensable si vous souhaitez conserver un minimum de contr&#244;le sur vos mod&#232;les et leur r&#233;silience &#224; la censure.</p><ul><li><p><a href="https://eurollm.io/">EuroLLM, LLMs made in Europe</a></p></li><li><p><a href="https://github.com/nvk/llm-wiki">LLM-wiki, a LLM-compiled knowledge base for any agent</a> - Github</p></li><li><p><a href="https://github.com/gonka-ai/gonka/">Gonka, a decentralized AI architecture</a> - Github</p></li><li><p><a href="https://boristane.com/blog/how-i-use-claude-code/">Using Claude Code by building a detailed plan</a></p></li><li><p><a href="https://dev.to/aws/stop-ai-agent-hallucinations-4-essential-techniques-2i94">Stop agents hallucinations</a> - dev.to</p></li><li><p><a href="https://github.com/VectifyAI/PageIndex">PageIndex, a vectorless, reasoning-based RAG</a> - Github</p></li></ul><div><hr></div><h1>PHP</h1><p><strong>O</strong>n revient au classique avec PHP et notamment un article sur l&#8217;&#233;cosyst&#232;me actuel de l&#8217;IA (au sens large du terme) au sein de PHP, il y a &#233;norm&#233;ment de librairies d&#233;j&#224; existantes dont quelques-unes que j&#8217;ai d&#233;couvertes, d&#8217;autres que je connaissais d&#233;j&#224; (je pense notamment &#224; TransformersPhp - un must si vous voulez des mod&#232;les locaux directement dans vos application Symfony / Laravel), l&#8217;article explique tr&#232;s bien que PHP est d&#233;j&#224; pr&#234;t pour l&#8217;IA mais qu&#8217;il continue &#224; souffrir des poncifs habituels, un &#8220;sacr&#233; g&#226;chis&#8221; quand on connait les statistiques du d&#233;ploiement de PHP &#224; travers le monde.</p><p>Je continue avec PicklePantherBundle, un bundle Symfony permettant d&#8217;impl&#233;menter &#8220;l&#8217;approche Behat&#8221; au sein de Panther (en pratique, la syntaxe me rappelle furieusement l&#8217;outil de test E2E d&#233;velopp&#233; par Blackfire il y a quelques ann&#233;es), j&#8217;avoue ne pas &#234;tre fan de la syntaxe YAML (parce que je n&#8217;aime pas cette syntaxe, c&#8217;est tout, rien de personnel avec le bundle) mais voir des extensions &#224; Panther fait toujours plaisir et permet de maintenir la librairie en vie (m&#234;me si Playwright a secou&#233; pas mal d&#8217;arbres dans l&#8217;&#233;cosyst&#232;me).</p><p>Dernier petit sujet avec de l&#8217;optimisation (j&#8217;ai lu l&#8217;article il y a une semaine et j&#8217;aurais pu le partager la semaine derni&#232;re mais la veille &#233;tait suffisamment charg&#233;e) concernant PHP-FPM et notamment sa mont&#233;e en charge, indispensable si vous voyez r&#233;guli&#232;rement vos infrastructures tomber apr&#232;s des milliers de requ&#234;tes.</p><ul><li><p><a href="https://linuxblog.io/php-fpm-tuning-using-pm-static-max-performance/">Tuning PHP-FPM for maximum performances</a></p></li><li><p><a href="https://github.com/Amoifr/pickle-panther-bundle">PicklePantherBundle, a YAML-driven E2E testing engine on top of Panther</a> - Github</p></li><li><p><a href="https://zadroweb.com/blog/php-ai-agents/">PHP is already built for the AI era</a></p></li><li><p><a href="https://github.com/Kocal/BiomeJsBundle">A Symfony Bundle to use Biome.js as a frontend linter</a> - Github</p></li><li><p><a href="https://blog.phalcon.io/post/phalcon-v6-0-0alpha1-released">Phalcon is now a &#8220;full PHP&#8221; framework</a></p></li><li><p><a href="https://www.exakat.io/php-and-ai-what-actually-exists-at-the-language-level/">The current state of AI at language level</a></p></li><li><p><a href="https://github.com/symfony/webpack-encore/releases/tag/v7.0.0">Webpack Encore 7 is out</a> - Github</p></li><li><p><a href="https://docs.volt-test.com/blog/php-stress-testing-tool/">Stress testing PHP applications using Volt Test</a></p></li><li><p><a href="https://github.com/php-opcua">PHP OPC UA, a pure implementation of the OPC UA protocol stack</a> - Github</p></li></ul><div><hr></div><h1>Tools</h1><p><strong>O</strong>n finalise cette veille avec les outils et notamment FanControl, un utilitaire d&#233;couvert par hasard sur X il y a une semaine (si j&#8217;ai pas de b&#234;tises) qui permet de se passer des solutions &#8220;closed source&#8221; (et par cascade, propri&#233;taires) pour le contr&#244;le de vos ventilateurs (GPU inclus), je suis en cours de tests sur, j&#8217;aurais donc probablement un avis &#224; formuler prochainement, si vous avez du temps, c&#8217;est &#224; tester (et c&#8217;est gratuit).</p><ul><li><p><a href="https://github.com/tobi/qmd">qmd, a mini CLI search engine for your documents and others</a> - Github</p></li><li><p><a href="https://github.com/Rem0o/FanControl.Releases">FanControl, a fan controller for your computer (Windows only)</a> - Github</p></li><li><p><a href="https://github.com/originalankur/maptoposter">MapToPoster, transform your favorite cities onto minimalist designs</a> - Github</p></li></ul><p>Voici donc pour la veille de cette semaine, comme d&#8217;habitude, n&#8217;h&#233;sitez pas &#224; laisser un commentaire / like / partager si le coeur vous en dit, de mon c&#244;t&#233;, je vous donne rendez-vous d&#232;s la semaine prochaine pour une nouvelle veille.</p><p>D&#8217;ici-l&#224;, passez un excellent weekend ainsi qu&#8217;une tr&#232;s bonne semaine.</p><p>A vendredi.</p>]]></content:encoded></item></channel></rss>