I know a lot of people in the West are baffled by the Mbappé-Mao storm in China - which in itself shows that Mbappé's ignorance doesn't come out of nowhere...
To understand it, here's the best analogy I could think of: imagine the biggest Chinese star in the NBA, with sponsorship deals all over the US. One day he's asked for his favorite meme of himself. He giggles: "the one where they put my face on some guy, I think he was an American slave owner." When the "some guy" in question is George Washington.
Whatever you think of the label, you know exactly how that would land in the US: an ungrateful foreigner getting rich off Americans, laughing at the country's founding father in an offensive way - at least to a big share of the population - without even knowing his name.
A founding father is the story a nation tells itself about where it comes from. Laugh at him as a foreigner while cashing in on that nation's consumers, and that's not going to go well in any country.
The label is also extremely un-nuanced: yes, all important historical figures have their share of darkness and Washington and Mao are no exception.
China actually recognizes this: the official line is that he was "70% right and 30% wrong," and the Party's 1981 resolution condemned how his "personal arbitrariness gradually undermined democratic centralism in Party life and the personality cult grew graver and graver" (https://t.co/gH2rs1FZqW)
Same for Washington: it's officially acknowledged too. Or at least it was until this year: in January, the National Park Service tore down an exhibit in Philadelphia about the people Washington enslaved there, under a Trump executive order targeting content that "inappropriately disparage[s]" Americans (https://t.co/MRLTnnllfq). Which goes to show China isn't uniquely touchy about its founding father...
O milei padece de disonancia cognitiva o practica la deshonestidad de la “geopolítica de patada en el culo zurdo” selectiva.
En Madrid grita desde el atril que irá a cualquier lugar del mundo a “enfrentar” a los liberticidas; pidió “sacar a los socialistas a patadas” y luego celebró que “no hay nada más divertido que patear culos zurdos”.
Olvidemos por un rato que se lo eligió para gobernar un país, no para largarse como cruzado desaforado por un mundo que no lo eligió como su salvador. Bien, entonces veamos qué implica ser zurdo y apliquemos su propia vara, sobre el tamaño del Estado.
España registra un gasto público equivalente a alrededor del 45,5% del PIB. Pero Finlandia ronda el 57,7%; la Francia de Macron, el 57,3%; Austria, el 55,2%; la Italia de su querida Meloni, el 50,4%; y Suecia, el 50%.
Ya sé lo que alguno dirá que el gasto público total también incluye defensa, intereses, justicia, infraestructura y administración. Perfecto, entonces miremos únicamente protección social y salud.
España destina a esas dos funciones el 25,2% del PIB. Finlandia, el 34,2%; Francia, el 32,6%; Austria, el 32,3%; Italia, gobernada por Meloni, el 27,9%; y Suecia, el 26,4%.
Es decir, incluso cuando sacamos Defensa y buena parte del resto del Estado, la Italia de Meloni sigue gastando socialmente más que la España de Sánchez.
Entonces, ¿cómo continúa la gira mundial contra los liberticidas?
¿Irá al Elíseo de nuevo a explicarle a Macron, con la misma delicadeza que reservó para Sánchez, que Francia tiene un Estado bastante más grande que España y se olvidó mencionar la última vez?.
¿Viajará a Roma para “sacar a patadas” a Meloni, o el gasto deja de ser socialista cuando lo ejecuta una amiga o ser miembro del club otorga dispensa doctrinal?
¿Aterrizará en Suecia o Austria cuyos gobiernos gastan más que el español y los aconsejará con su verba tan educada con la que averguenza a cualquier persona con un mínimo de educación y recato?
Trump merece un capítulo aparte: Estados Unidos gasta menos que España, pero el FMI calcula un déficit del 7 al 7,5% del PIB, una deuda encaminada a superar el 140% y una política arancelaria abiertamente proteccionista. ¿Lo acusará de “degenerado fiscal” y le explicará que el proteccionismo es corrupción, como sostuvo ayer? ¿O en Washington la Escuela Austríaca entra en receso?. Vale para el proteccionismo comercial que predica Vox y Abascal también.
Y después queda China. Su Constitución establece literalmente un Estado socialista dirigido por el Partido Comunista. ¿Cuándo parte a Pekín para enfrentar a Xi? Porque cuando lo tuvo delante no hubo patadas; más bien le agradeció el “valioso apoyo” financiero chino y ofreció cooperación.
La explicación es bastante más sencilla de lo que parece; no existe una cruzada universal contra el estatismo como tampoco existen países que estén dispuestos a invitar y recibir a semejante ignorante y maleducado.
A Sánchez o Lula les corresponden insultos. A Macron, reuniones cordiales. A Meloni y antes Orbán, amistad. A Trump, veneración. A Xi, pragmatismo financiero.
El gasto público no convierte automáticamente a un gobierno en izquierdista, es importante también el déficit, la deuda, las regulaciones, el proteccionismo, el respeto por la propiedad privada y la calidad institucional. Pero fue Milei quien convirtió el tamaño del Estado en prueba moral contra sus adversarios. Si ésa es la vara, debe aplicarla también a sus amigos.
Un liberal puede criticar cualquier política extranjera. Lo que no puede hacer con honestidad intelectual es cambiar de doctrina según quién esté sentado enfrente.
Y existe una cuestión todavía más elemental para estas giras barnizadas como oficiales: a los gobernantes de otros países los eligen y los remueven sus propios ciudadanos. Un presidente argentino no tiene mandato para ir recorriendo el mundo “pateando” culos adversarios insultando a media sociedad del país que lo recibe, ni interviniendo en campañas ajenas.
Mucho menos mientras el vocero sostiene que todos esos viajes son oficiales y la factura la pagan los argentinos.
PS: busquen las estadísticas en Eurostat, ahi tienen el detalle de las cifras citadas.
i'm begging you:
- sell boring consumable products to boring demographics
- sell dog supplements. cleaning refills. coffee pods. teeth whitening strips
- things that run out and need replacing
- sell them to women 45-65 who don't know what temu is
- charge $75/month on subscription
- pre-check the subscribe box
- the product runs out. the reorder is automatic. the customer doesn't think about it
- do this with 15-20 products across 3-5 niches
- each product does $2-5K/month. not impressive alone
- 20 products at $3K average = $60K/month
- one product dies = 5% revenue dip. not a crisis
- the one-product brand loses a product and loses 100%
- check your phone twice a day
- "I sell boring consumables to boring people" doesn't make a good tweet. it makes a good bank statement
1) why consumables?
- the product creates its own demand. it runs out. the reorder trigger is biological not emotional
- coffee runs out. supplements run out. cleaning products run out. phone cases don't
- the empty container is the retargeting. no email needed. no ad needed. the empty bottle does the selling
- LTV on consumables: $280-1,440/customer. LTV on one-time products: $40. same acquisition cost. 7-36x more revenue per customer
2) why 15-20 products instead of 1?
- every product dies eventually. 3-6 months at peak before competition, creative fatigue, or seasonal shifts kill it
- 1 product dying = 100% revenue gone. total crisis
- 1 of 20 products dying = 5% revenue dip. barely noticeable. replace it next weekend
- the catalog gets stronger every month. only winners survive. losers get replaced. the system cleans itself
3) why this demographic?
- holds 53% of all US household wealth. the 18-34 group holds 5%
- 400,000 operators fight over the 5%. you're selling to the 53%
- she subscribes and stays. average 8-12 months. churn under 12%
- she doesn't compare prices. doesn't check alternatives. doesn't return products
- acquiring her costs $15-20. she pays back $280-1,440 over the subscription. nobody targets her because selling menopause supplements doesn't look good on instagram
4) how do you find 20 consumable products?
- tiktok comments in health, pet, kitchen, skincare, cleaning
- look for products people buy repeatedly not products people buy once
- "I buy this every month" in the comments = better signal than "where do I buy this"
- the product that gets reordered is worth 7-36x more than the product that goes viral once
5) why does nobody talk about this?
- because "I run 20 boring subscriptions to middle-aged women on facebook" gets zero engagement on twitter
- the operators doing $60K/month on catalog consumables don't post about it because there's nothing exciting to screenshot
- the operators posting $50K/month screenshots are usually selling one viral product that dies in 3 months
- the catalog operator makes less noise and more money over 12 months than the viral operator makes in their best month
boring compounds. viral peaks and dies. pick which one you want
[ full system is in a 36-page playbook. comment "BORING" and I'll send it. must be following ]
(I built the AI tool that does the page part - paste any aliexpress link, full page in 60 seconds → link in bio)
@arquiguru Ese video lo grabó mi novia desde nuestro balcón, nos reímos un rato cuando lo vimos porque la letra de la canción hacía match con el mood de la escena, demasiado cine, le dije que lo publique y reventó 🤣
A few people asked how I made the isometric objects and cities, so I packaged it into a Claude skill.
Describe an object, device or building, and ISO-GLOW creates a single interactive HTML file.
Try it out yourself!
https://t.co/ncC0vK8B3h
Cada video que sale del incidente sancor es mejor que el anterior
El chiqui tapia hizo un atenrado terrorista en vivo
Libero un riesgo cognitivo asi nomas, ni lovecraft te hacia una historia asi
i spent 70+ hours with Opus 5.5 inside Claude Code... here's how to get the best out of it:
- set effort to medium at all times, more effort never gave me better outputs
- keep your context window under 400k tokens
- only use a global claude․md, let the code and your knowledge base do the rest
- use git worktrees and keep one main chat as the router
- prompt the router to spawn each project's thread in its own worktree, versioning gets easy
- use subagents a lot, they keep your main context window clean
- you share the vision and the end goal, the workers do the research and coding
- skip plan mode, use /grill-me or the Compound engineering plugins instead
- kick off new projects with heavy research and a pile of inspirations
- then trust Opus overnight, with the right references it can build literally anything
- disable 90% of your skills, you probably don't need them anymore
- have Opus build an artifact that maps the full scope of each project
- it updates the map at every milestone, so you always see what's done and what's left
Hot take: evals are the #1 thing keeping companies from becoming AI-native.
If you want to understand evals, read this post & then the accompanying guide.
Tweaking prompts and trying a few inputs by hand only gets you so far; evals are what get your agents ready for deployment and keep them working once they're live.
First, some definitions...
- An eval: a grader applied to a trace.
- A grader: anything that scores how well the agent did, from a simple code check to an LLM acting as a judge.
- A trace: the full record of one agent run: the input, every step it took, and the output. For example, a customer asks "Where is my refund?" The agent calls search_orders, then issue_refund, and replies "Refund sent, 3–5 days."
Next, let's dive into graders...
Graders come in a few families, and most real setups run several on the same trace.
1) Deterministic checks. Plain code, no model involved. Did the agent call issue_refund exactly once, with the same order ID it looked up? A double refund or a refund on the wrong order fails instantly.
2) Reference-answer graders. Compare the output to a reference answer stored on the case, by exact match or with an LLM judge checking whether the meaning matches.
3) Rubric graders. An LLM judge scores the output against criteria, with no reference answer needed. Is the reply concise? Does it confirm what happened?
4) Trajectory graders. These grade the steps the agent took, not just its final answer. Did it look up the order before issuing the refund? A run that calls issue_refund without checking the order first fails even if the final reply reads perfectly.
What are online vs. offline evals?
Online evals watch production. Offline evals run in your lab: on demand or in CI, before anything ships.
- You need both. Online without offline means you can see problems but can't fix them safely. - Offline without online means you're testing the scenarios you imagined, not the ones your users actually create.
- What connects the two is promotion, turning a bad production trace into an offline case.
Next, eval-driven development...
Eval-driven development is test-driven development for agent behavior.
1) Define the case and its pass criteria.
2) Run it and watch it fail.
3) Change the agent.
4) Run it again. Repeat steps 3 and 4 until it passes.
5) Run the full dataset, not just this case, to catch regressions.
Run each case several times (k) and set the bar explicitly: all-of-k for high-stakes behavior where a single failure is unacceptable, or n-of-k, like 2 of 3, where some variance is acceptable.
Put together, online evals find a problem, promotion turns it into a case, the loop fixes it offline, and the growing dataset makes sure it stays fixed.
Read our full evals guide here:
🚨ESTO ES JODIDAMENTE INCREÍBLE
un tío acaba de enseñar cómo Claude le gestiona todas las redes sociales con una sola conexión MCP.
no es una demo. son sus cuentas reales.
le da los clips. Claude saca la transcripción con Whisper, escribe títulos, descripciones, hashtags y lo programa todo en 5 plataformas.
una semana de contenido. 7 minutos.
después le pide un informe de analytics. Claude lee 30 días de datos y le dice en qué doblar la apuesta.
después le dice que gestione el inbox. 50 DMs y 30 comentarios. respuestas redactadas. lo urgente marcado.
y lo dejó montado como rutina automática. cada día.
se llama Zernio. setup de 5 minutos. sin código.
os dejo el video abajo.