Os comparto mi último post: por qué tus agentes de IA no aprenden de sus errores y cómo arreglarlo.
👉 iamdgarcia.medium.com/your-a…
16
Nuevo post ya disponible!
Demo y código para crear un asistente que coja las llamadas de tu negocio y te gestione las citas
iamdgarcia.substack.com/p/la…
141
Un punto de vista muy interesante sobre lo que está pasando con Google
A year ago, Microsoft, Alphabet and Meta generated $60.7 billion of free cash flow in a single quarter. This summer, $7.4 billion.
They didn't sell less. Revenue grew 19% at Microsoft, 25% at Alphabet and 26% at Meta. What changed is the bill: $125.6 billion in three months for data centers, chips and servers, double what they spent a year earlier. Meta put 54 cents of every dollar of revenue into capex. Alphabet spent $9 billion more than it took in, its second negative quarter in a row, and this morning its stock fell 12%. If it closes down more than 9.77%, it will be its worst day after earnings since going public in 2004.
The detail that worries me most is in Google's search numbers. Search ads grew 12% and still missed expectations. Philipp Schindler, Alphabet's chief business officer, said part of the reason is that AI Mode makes less money per search. So Google is spending record sums to build a product that earns less than the one it replaces.
For twenty years these were the best businesses ever invented. You wrote the code once, sold it to billions of people, and almost every new dollar of revenue fell to the bottom line. AI doesn't work like that. Every answer runs on a chip someone had to buy and on electricity someone has to pay for. Big Tech is turning into something closer to a utility: huge upfront costs, long paybacks, and profits that depend on keeping the machines full.
That can still be a great business. But it is a different business, and investors are only starting to price it that way.
The old moat was code. The new one is cash.
24
Si tenéis por casa algún portátil viejo, GPUs antiguas o un Mac de hace mil años: ya podéis ejecutar un modelo capaz de automatizaros la vida en cero coma
Tool-calling y coherencia en iPads, iPhones, relojes y hasta en el frigorífico
Barra libre para todos
Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence.
> 6 model instances, across text, audio, vision, and omni modalities, generative and decision models.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding
> 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages.
> 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language.
> 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input.
> fine-tune on 1 GPU.
> a new tri-hybrid architecture
> architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs.
> 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters.
> 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1 on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒
Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device.
1/ 🧵
1
2
74
iamdgarcia.substack.com/p/55…
Más no siempre es mejor.
En el último episodio del podcast hablamos sobre por qué la estructura de agente + skills suele ganar a la estructura multiagente
93
Justo de esto hablamos en el último artículo
Link aquí iamdgarcia.medium.com/anthro…
Anthropic is now recommending that companies REMOVE unnecessary subagents from their AI systems.
This is a pretty significant shift in agent architecture, with their team finding that a single agent equipped with skills can frequently outperform complex multi-agent systems, and one customer seeing roughly 2x faster execution.
The emerging rules of thumb:
→ Skills: Teach the agent HOW to do something, including procedures, expertise, and instructions.
→ Tools / MCP: Give the agent the ability to DO things through APIs, actions, and live data.
→ Subagents: Use for parallel work, isolated context, or tasks a single agent struggles with.
The interesting part is progressive disclosure. Instead of spawning specialized agents or stuffing everything into a giant prompt, one agent discovers and loads skills and tools as needed.
Their advice is to fold existing subagents into the main loop, turn their instructions into skills, expose their actions as tools, and run evals to see if performance holds up.
As models have gotten smarter, improving agent architecture is becoming less about what you can add and more about what you can delete.
claude.com/resources/article…
25
Un controlador de decisiones subió el coste de un agente un 19% por intento y el acierto se quedó exactamente igual: 73,0% contra 73,0%. Cero puntos. Y no es un caso raro, el experimento es placebo-controlled.
1
10
Me gustó este paper porque confirma con experimento controlado lo que vimos en nuestras propias trazas: el coste por intento no es el coste por éxito entregado. Puedes encadenar llamadas baratas y aun así entregar peor.
1
1
Por eso medimos contained, abandoned y handoff en vez de intentos: github.com/iamdgarcia/trace2…
¿Cuánto de tu factura es trabajo ya correcto que se verificó dos veces?
3
estamos en directo hablando de agentes conversacionales open.substack.com/live-strea…
107
Daniel García retweeted
Voice agents fail in the gaps between turns, not in the words. So I built 4 instruments you can
open in a browser tab and watch it happen: endpointing live, word timing, turn KPIs, latency.
No signup, everything runs client-side.
iamdgarcia.medium.com/watchi…
Voice agents fail in the gaps between turns, not in the words. So I built 4 instruments you can
open in a browser tab and watch it happen: endpointing live, word timing, turn KPIs, latency.
No signup, everything runs client-side.
iamdgarcia.medium.com/watchi…
Anthropic acaba de publicar un blog donde explican cómo gestionan todo lo relacionado con revisión de código generado con agentes
Muy interesante
Claude writes about 90% of Claude Code's code.
From June to September, merged PRs more than doubled while the share reverted fell by more than 40%.
We wrote about how we changed code review to get there, and how to adopt it:
claude.dev/blog/solving-code…
18
Os comparto el último vídeo de YouTube donde hablamos sobre la diferencia fundamental entre los modelos System 1 y los LLMs
youtu.be/TWPKpRjYmNw?is=iWAn…
211
Is it possible to replicate a Jev-like model using a llm?
That is what I explored in my latest article
iamdgarcia.medium.com/build-…
39
new article available! Let's see how to teach jev to play flappy bird!
iamdgarcia.medium.com/i-buil…
57






