<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Blog transformation IA, agents et intégration .NET | GiLabs</title>
    <link>https://gilabs.fr/fr/blog/</link>
    <description>Retours d&apos;expérience et méthode sur la transformation IA des PME, ETI et cabinets : agents IA, intégration .NET, automatisation de process, cas concrets.</description>
    <language>fr-FR</language>
    <lastBuildDate>Thu, 17 Sep 2026 00:00:00 GMT</lastBuildDate>
    <atom:link href="https://gilabs.fr/fr/rss.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Ce qui fait tenir un pipeline documentaire en production</title>
      <link>https://gilabs.fr/fr/blog/ce-qui-tient-en-production/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/ce-qui-tient-en-production/</guid>
      <pubDate>Thu, 17 Sep 2026 00:00:00 GMT</pubDate>
      <description>Trois décisions d&apos;ingénierie séparent un POC IA d&apos;un système qui tourne tous les jours : revue humaine, boucle de correction, briques d&apos;extraction interchangeables.</description>
      <category>IDP</category>
      <category>Architecture</category>
      <category>Production</category>
      <category>Transformation IA</category>
    </item>
    <item>
      <title>On a déployé Copilot, et rien n&apos;a bougé dans les comptes</title>
      <link>https://gilabs.fr/fr/blog/copilot-deploye-rien-na-change/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/copilot-deploye-rien-na-change/</guid>
      <pubDate>Tue, 15 Sep 2026 00:00:00 GMT</pubDate>
      <description>Copilot accélère la personne. Le coût d&apos;un process est fait du temps qui s&apos;écoule, pas du temps qu&apos;on y travaille. Voici comment mesurer les deux.</description>
      <category>Copilot</category>
      <category>Transformation IA</category>
      <category>PME</category>
      <category>Process</category>
    </item>
    <item>
      <title>Un diagnostic IA de 30 jours, une journée de votre temps</title>
      <link>https://gilabs.fr/fr/blog/ce-que-le-diagnostic-demande-aux-equipes/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/ce-que-le-diagnostic-demande-aux-equipes/</guid>
      <pubDate>Tue, 08 Sep 2026 00:00:00 GMT</pubDate>
      <description>Le blocage avant de lancer un diagnostic IA n&apos;est presque jamais le prix, c&apos;est la charge sur les équipes. Voici le chiffre exact : trois jalons, quelques entretiens, une décision.</description>
      <category>Transformation IA</category>
      <category>PME</category>
      <category>Méthode</category>
      <category>Diagnostic</category>
    </item>
    <item>
      <title>Le POC IA qui ne passera jamais en prod tel quel</title>
      <link>https://gilabs.fr/fr/blog/industrialiser-poc-python-dotnet/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/industrialiser-poc-python-dotnet/</guid>
      <pubDate>Wed, 12 Aug 2026 00:00:00 GMT</pubDate>
      <description>Un POC Python qui convainc en démo mais reste bloqué six mois n&apos;a pas besoin d&apos;être réécrit en C#. Il a besoin d&apos;être exposé derrière une API et intégré aux règles du reste du SI.</description>
      <category>.NET</category>
      <category>Architecture logicielle</category>
      <category>Transformation IA</category>
      <category>Python</category>
    </item>
    <item>
      <title>Qui a dit quoi, sans reconnaissance vocale : la diarisation par métadonnée</title>
      <link>https://gilabs.fr/fr/blog/diarisation-sans-reconnaissance-vocale/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/diarisation-sans-reconnaissance-vocale/</guid>
      <pubDate>Mon, 10 Aug 2026 00:00:00 GMT</pubDate>
      <description>Attribuer chaque phrase d&apos;un compte-rendu de réunion à la bonne personne sans créer de donnée biométrique. Trois pistes comparées, une seule ne crée aucune empreinte vocale.</description>
      <category>RGPD</category>
      <category>Intelligence artificielle</category>
      <category>Réunions</category>
      <category>.NET</category>
    </item>
    <item>
      <title>Le pivot qui a sauvé un projet d&apos;IA vocale : penser terrain avant de coder</title>
      <link>https://gilabs.fr/fr/blog/pivot-capture-teams/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/pivot-capture-teams/</guid>
      <pubDate>Fri, 31 Jul 2026 00:00:00 GMT</pubDate>
      <description>Un projet prévoyait qu&apos;un technicien appelle un assistant automatisé pour dicter son estimation de chantier. Quatre obstacles bien concrets ont montré que cette idée ne tenait pas sur le terrain.</description>
      <category>Intelligence artificielle</category>
      <category>Microsoft Teams</category>
      <category>Terrain</category>
    </item>
    <item>
      <title>La voix des salariés ne doit jamais sortir d&apos;Europe : une règle posée avant la technologie</title>
      <link>https://gilabs.fr/fr/blog/whisper-onprem-souverainete/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/whisper-onprem-souverainete/</guid>
      <pubDate>Fri, 31 Jul 2026 00:00:00 GMT</pubDate>
      <description>Pour produire un compte-rendu de réunion fiable, une direction a d&apos;abord tranché une question de principe : où partent les enregistrements de voix ? Trois solutions comparées, une seule retenue.</description>
      <category>Protection des données</category>
      <category>Intelligence artificielle</category>
      <category>Réunions</category>
    </item>
    <item>
      <title>RAG sur un SI .NET existant : trois décisions qui changent tout</title>
      <link>https://gilabs.fr/fr/blog/rag-app-dotnet-existante/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/rag-app-dotnet-existante/</guid>
      <pubDate>Tue, 28 Jul 2026 00:00:00 GMT</pubDate>
      <description>Brancher un RAG sur un patrimoine applicatif .NET existant n&apos;est pas d&apos;abord un choix de modèle. C&apos;est une question de stockage vectoriel, de découpage documentaire et d&apos;orchestration qui respecte l&apos;existant.</description>
      <category>RAG</category>
      <category>.NET</category>
      <category>Semantic Kernel</category>
      <category>Architecture</category>
    </item>
    <item>
      <title>Diagnostic IA 30 jours : pourquoi j&apos;ai packagé mon offre en trois livrables fixes</title>
      <link>https://gilabs.fr/fr/blog/diagnostic-ia-30-jours-packaging/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/diagnostic-ia-30-jours-packaging/</guid>
      <pubDate>Mon, 27 Jul 2026 00:00:00 GMT</pubDate>
      <description>Un forfait avec trois livrables écrits, pas un audit qui s&apos;arrête au rapport : pourquoi une offre de conseil IA packagée se décide plus vite qu&apos;une conversation ouverte.</description>
      <category>Diagnostic IA</category>
      <category>Transformation IA</category>
      <category>PME</category>
      <category>Offre</category>
    </item>
    <item>
      <title>Recruter un dev .NET et un data scientist en même temps : un seul besoin, pas deux</title>
      <link>https://gilabs.fr/fr/blog/pont-dotnet-ia-recrutement/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/pont-dotnet-ia-recrutement/</guid>
      <pubDate>Wed, 22 Jul 2026 00:00:00 GMT</pubDate>
      <description>Une entreprise qui recrute simultanément un développeur .NET et un data scientist envoie un signal clair : elle ne sait pas encore qu&apos;elle cherche une seule personne.</description>
      <category>Recrutement</category>
      <category>.NET</category>
      <category>Transformation IA</category>
      <category>SI existant</category>
    </item>
    <item>
      <title>L&apos;IA propose, le déterministe dispose : le pattern qui fait tenir une IA en production</title>
      <link>https://gilabs.fr/fr/blog/ia-propose-deterministe-dispose/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/ia-propose-deterministe-dispose/</guid>
      <pubDate>Wed, 15 Jul 2026 00:00:00 GMT</pubDate>
      <description>Une IA en production n&apos;est pas une IA qui décide seule. C&apos;est une IA qui propose, à un système qui a le droit de dire non. Le pattern d&apos;architecture en trois couches.</description>
      <category>Architecture</category>
      <category>.NET</category>
      <category>Agents IA</category>
      <category>Fiabilité</category>
    </item>
    <item>
      <title>Votre problème n&apos;est pas (encore) un problème d&apos;IA</title>
      <link>https://gilabs.fr/fr/blog/socle-data-avant-ia/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/socle-data-avant-ia/</guid>
      <pubDate>Fri, 10 Jul 2026 00:00:00 GMT</pubDate>
      <description>« On veut de l&apos;IA sur notre documentation. Elle est où, votre documentation ? Dans des classeurs. » Le socle de données passe avant l&apos;IA, et c&apos;est une bonne nouvelle.</description>
      <category>Maturité data</category>
      <category>Transformation IA</category>
      <category>PME</category>
      <category>Diagnostic</category>
    </item>
    <item>
      <title>Concierge d&apos;abord, code ensuite : pourquoi j&apos;ai passé des semaines à ne rien automatiser</title>
      <link>https://gilabs.fr/fr/blog/concierge-dabord-code-ensuite/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/concierge-dabord-code-ensuite/</guid>
      <pubDate>Wed, 08 Jul 2026 00:00:00 GMT</pubDate>
      <description>Avant d&apos;écrire la première ligne de mon agent de prospection, j&apos;ai fait le travail à la main, longtemps. C&apos;est cette étape, et non le code, qui fait la différence.</description>
      <category>Méthode</category>
      <category>n8n</category>
      <category>Automatisation</category>
      <category>Cartographie process</category>
    </item>
    <item>
      <title>« C&apos;est cher » : un problème de catégorie, pas de prix</title>
      <link>https://gilabs.fr/fr/blog/outil-vs-transformation-categorie/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/outil-vs-transformation-categorie/</guid>
      <pubDate>Tue, 07 Jul 2026 00:00:00 GMT</pubDate>
      <description>Un outil IA est une dépense. Une transformation IA est un investissement. Confondre les deux grilles de lecture explique la plupart des « c&apos;est trop cher » entendus en PME.</description>
      <category>Transformation IA</category>
      <category>ROI</category>
      <category>PME</category>
      <category>Pricing</category>
    </item>
    <item>
      <title>Étude METR 2025 : les développeurs se croyaient 20 % plus rapides, ils étaient 19 % plus lents</title>
      <link>https://gilabs.fr/fr/blog/metr-productivite-ia/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/metr-productivite-ia/</guid>
      <pubDate>Thu, 02 Jul 2026 00:00:00 GMT</pubDate>
      <description>Un essai contrôlé sur des développeurs expérimentés montre un écart frappant entre ressenti et mesure. Le gain de l&apos;IA n&apos;est ni automatique ni gratuit : il se mesure.</description>
      <category>Productivité</category>
      <category>ROI</category>
      <category>Mesure</category>
      <category>Recherche</category>
    </item>
    <item>
      <title>L&apos;usine logicielle IA de Garry Tan : la leçon n&apos;est pas dans les outils</title>
      <link>https://gilabs.fr/fr/blog/usine-logicielle-ia/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/usine-logicielle-ia/</guid>
      <pubDate>Wed, 01 Jul 2026 00:00:00 GMT</pubDate>
      <description>Le président de Y Combinator a publié sa méthode pour « shipper comme une équipe de vingt ». La vraie leçon n&apos;est pas dans les 30 outils qu&apos;il donne, mais dans le process qu&apos;ils appliquent.</description>
      <category>Industrialisation</category>
      <category>Méthode</category>
      <category>IA générative</category>
    </item>
    <item>
      <title>« On a déjà Copilot, pourquoi vous ? » : la question à laquelle il faut répondre</title>
      <link>https://gilabs.fr/fr/blog/objection-copilot/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/objection-copilot/</guid>
      <pubDate>Wed, 24 Jun 2026 00:00:00 GMT</pubDate>
      <description>L&apos;objection numéro un des dirigeants sous environnement Microsoft. Trois choses qu&apos;une licence Copilot n&apos;achète pas, et la vraie question à se poser avant d&apos;y répondre.</description>
      <category>Microsoft Copilot</category>
      <category>Transformation IA</category>
      <category>Conformité</category>
      <category>PME</category>
    </item>
    <item>
      <title>J&apos;ai construit un agent IA pour ma prospection : voici où il se trompe</title>
      <link>https://gilabs.fr/fr/blog/agent-n8n-prospection/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/agent-n8n-prospection/</guid>
      <pubDate>Fri, 19 Jun 2026 00:00:00 GMT</pubDate>
      <description>Un agent n8n sourcait, enrichissait et notait mes prospects chaque matin. Ce qu&apos;il m&apos;a le mieux appris, c&apos;est un cas concret où une IA confiante peut être complètement fausse.</description>
      <category>n8n</category>
      <category>Automatisation</category>
      <category>IA générative</category>
      <category>Build in public</category>
    </item>
    <item>
      <title>Zéro validation humaine, et c&apos;est le bon choix : la chaîne IA de Convention Online</title>
      <link>https://gilabs.fr/fr/blog/convention-online-ia-autonome/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/convention-online-ia-autonome/</guid>
      <pubDate>Wed, 17 Jun 2026 00:00:00 GMT</pubDate>
      <description>Un pipeline exige une validation humaine à chaque étape. Un autre tourne toute la nuit sans personne. Comment décider où mettre le curseur, selon l&apos;enjeu métier réel.</description>
      <category>Automatisation</category>
      <category>IA générative</category>
      <category>Architecture</category>
      <category>SEO</category>
    </item>
    <item>
      <title>Un pipeline documentaire IA en production : ce qui le fait tenir au quotidien</title>
      <link>https://gilabs.fr/fr/blog/pipeline-idp-production/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/pipeline-idp-production/</guid>
      <pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate>
      <description>Des centaines de feuilles de temps scannées, saisies à la main, sans contrôle. Voici le pipeline IA que j&apos;ai conçu pour un acteur industriel, et ce qui compte vraiment en production.</description>
      <category>IDP</category>
      <category>Azure AI</category>
      <category>.NET</category>
      <category>Automatisation</category>
    </item>
    <item>
      <title>Pourquoi j&apos;ai choisi Semantic Kernel plutôt que Python pour l&apos;IA en .NET</title>
      <link>https://gilabs.fr/fr/blog/certification-az2005-semantic-kernel/</link>
      <guid isPermaLink="true">https://gilabs.fr/fr/blog/certification-az2005-semantic-kernel/</guid>
      <pubDate>Mon, 08 Jun 2026 00:00:00 GMT</pubDate>
      <description>Certifié Microsoft AZ-2005, j&apos;ai fait le choix d&apos;intégrer l&apos;IA dans l&apos;existant .NET plutôt que de tout réécrire en Python. Le raisonnement, et le pattern Semantic Kernel.</description>
      <category>Semantic Kernel</category>
      <category>.NET</category>
      <category>Azure OpenAI</category>
      <category>Certification</category>
    </item>
  </channel>
</rss>
