Skip to content
Mis Reglas Para Que Claude Code No Me Mienta episode artwork

Mar 10, 2026 57 min

Mis Reglas Para Que Claude Code No Me Mienta

from Codemancers - Inteligencia Artificial

¡Vótame en los Premios iVoox 2026! Tu IA escribe código que parece perfecto, los tests pasan en verde... y es todo inventado. Estas son nuestras reglas para pillarlo. En este episodio desgranamos los 4 tipos de alucinaciones que los LLMs cometen al programar según el estudio CodeHallu (Universidad de Hong Kong) y compartimos las defensas reales que usamos a diario: conectar Claude Code con Notebook LM vía MCP para delegar la investigación, Git hooks con pre-commit (Ruff) y pre-push (pytest) como capas de verificación automática, hooks propios de Claude Code para bloquear atajos prohibidos, y una skill llamada "improve" que lanza 3 subagentes (Kent Beck, Martin Fowler, Mike Acton) a revisar tu código desde arquitectura, simplicidad y rendimiento. También hablamos de los nuevos comandos /loop y /simplify de Anthropic, la herramienta Z3 para instrumentar apps en Mac, skills.sh como ecosistema de skills, y por qué la gente está usando Claude Code para cosas que no son programación. ⏱️ Timestamps: 0:00 Introducción 0:20 El concilio de expertos: skill de panel de debate con IA 2:31 Claude Code en 2026: por qué ya no usamos otra cosa 4:09 Conectar Claude Code con Notebook LM (MCP) 8:52 Demo en vivo: notebook de investigación sobre agentes IA 10:18 Claude Code para cosas que no son código (Remotion, marketing, Meta Ads) 12:01 CodeHallu: los 4 tipos de alucinaciones al programar 13:23 Mapping: cuando cruza identificadores válidos 15:14 Naming: se inventa nombres plausibles (solo 39% válidos en GPT-4o) 16:10 Fantasma: campos inventados en APIs 18:20 House of Cards: el castillo de naipes de errores acumulados 22:18 Crear interfaces para máquinas, no forzar las nuestras (herramienta Z3) 23:04 Git hooks: pre-commit con Ruff y pre-push con pytest 30:43 Hooks de Claude Code: pillarlo con las manos en la masa 39:03 /loop y /simplify de Anthropic + cron para Claude Code 43:05 Skill "improve": 3 subagentes Jedi revisan tu código 48:03 Skills.sh: el ecosistema de skills para todos los agentes 51:44 Proyecto open source: administrador web para Ollama 53:07 Reflexión final: no evitar alucinaciones, detectarlas a tiempo ¿Cuál ha sido la peor alucinación que os ha colado un LLM programando? ¿Tests en verde con datos inventados? Contadlo en comentarios Recursos mencionados: - Skills.sh (ecosistema de skills): https://skills.sh - Ztrace (herramienta de instrumentación para Mac): https://github.com/frr149/ztrace - MCP de Notebook LM: https://www.linkedin.com/feed/update/urn:li:activity:7436077823548661787/ - Cron para Claude Code: https://github.com/frr149/claude-cron - Remotion (editor de vídeo en React): https://remotion.dev #ClaudeCode #IAProgramacion #CodingWithAI #LLM #Anthropic

Episode details come from the podcast's RSS feed.

Embed this episode

Ready to play

Mis Reglas Para Que Claude Code No Me Mienta

0:00 57:46

No transcript yet.

Requests help us decide what to transcribe next.

AI Daily News Podcast Really Easy AI AI Daily News: Your premier source for cutting-edge artificial intelligence updates! Dive into the world of machine learning, deep learning, and data science with our daily tech briefings. From neural networks to natural language processing, we cover groundbreaking AI research, innovative applications, and industry trends. Explore the latest in computer vision, robotics, autonomous systems, and the Internet of Things. Stay informed on AI ethics, machine learning algorithms, and the transformative impact of AI on business, healthcare, and society. https://www.youtube.com/@AINewsFresh JC Speaks Ai JC Tambeau Introducing "JC Speaks AI" - a captivating and thought-provoking podcast. This podcast is born out of my passion for Artificial Intelligence and a deeply personal mission. At the heart of this endeavour is my granddaughter, Abby, a brave 7-year-old battling lung disease and damage due to medical negligence. A routine visit to the emergency department at the age of three resulted in her right lung consolidating and collapsing, forever altering her future. The health system not only failed her once but twice when they refused to cover her future medical care - care that could improve her life, care that will inevitably be costly. Host JC brings you engaging conversations and insights at the intersection of Artificial Intelligence, health, and wellness. We delve into cutting-edge AI applications and discuss groundbreaking research, engaging with experts to demystify the rapidly evolving world of AI. From personalized medicine to mental health, and from ethical considerations to t McKinsey on AI McKinsey Analytics A podcast mini-series on artificial intelligence from McKinsey, featuring conversations with a wide range of experts in the field that explore how to apply AI to the benefit of your business, and society. The series includes a look at corporate AI trends and impact; interviews with AI experts; and how governments are approaching AI. Tech 24 FRANCE 24 English As the digital age charges on, what do we stand to gain, and what might we lose? Who's calling the shots and why, from artificial intelligence to data protection? No matter how much time you spend online, technology is changing your life. Find out how every Sunday at 3:09pm Paris time.

Frequently asked questions

How long is this episode of Codemancers - Inteligencia Artificial?

This episode is 57 minutes long.

When was this Codemancers - Inteligencia Artificial episode published?

This episode was published on March 10, 2026.

Can I download this Codemancers - Inteligencia Artificial episode?

Yes. Use the download control on the episode player to save the publisher-provided media file.