CLOSED retweeted
YOUR AI AGENT KEEPS GETTING BLOCKED. THE WEB THINKS IT’S A BOT.
you ask it to book a table, order groceries, fill in a form
it opens the site, hits a CAPTCHA, and says: “I was blocked. Can you finish this yourself?”
that’s the part nobody talks about. agents don’t fail because they’re not smart. they fail at the browser
websites run anti-bot systems that read hundreds of tiny signals to decide if a visitor is a person
they were built to stop scrapers and spam
today they stop your assistant too
Tilion is browser infrastructure built from the browser up to pass those checks
so the site loads like it would for a person, and the agent sees the real page instead of a CAPTCHA
Tilion:
→ #1 on the public leaderboards for agent browsers: speed, start-up time and getting through protected sites
→ a browser ready in about a tenth of a second
→ runs ten browsers or ten million, so a company can send out millions of web agents at once
and the setup is one sentence: “setup tilion.com”
you type it into your agent (Claude, Poke, OpenClaw, Hermes or whichever you use) and it browses through Tilion from then on
the web was built for people. agents need a browser that gets treated like one
agents keep hitting captchas and blocks on the web.
today we're launching tilion (@ycombinator): the world's fastest and stealthiest browser infra for agents.
#1 in stealth
#1 in agent action speed
#1 in start-up time
dont believe it? tell your agent:
setup tilion.com
CLOSED retweeted
my life after my robots stopped needing me for the captcha
agents keep hitting captchas and blocks on the web.
today we're launching tilion (@ycombinator): the world's fastest and stealthiest browser infra for agents.
#1 in stealth
#1 in agent action speed
#1 in start-up time
dont believe it? tell your agent:
setup tilion.com
CLOSED retweeted
Esto es una locura. Google acaba de sacar un curso gratis de 2 horas sobre Harness engineering completo, del primer paso al ultimo:
0:29 - Harness engineering desde cero
11:08 - tu primer agente, bien construido
28:08 - Eval engineering con un agente “Chief of Staff”
1:00:13 - Loop engineering
1:23:19 - un sistema de grafos que se mejora solo y nunca se detiene
He visto cursos de 500$ que no tienen el valor que tiene este video
CLOSED retweeted
a table with every number correct scored zero
the standard errors were just sitting in the same cell instead of on the row below
→ 8 parsers, 12 public benchmarks
→ same pages tilted 2 to 4 degrees, still easy to read
→ first place changed on 6 of 12
on 9 of 12 the winner's lead was smaller than the median drop from the tilt
look at the chart, then run your own docs through parsejudge.com
Last week, I said every parser is #1 on its own benchmark. So we tested the benchmarks.
We ran 8 parsers across 12 benchmarks, then tilted the pages by 2–4°. 6 benchmarks got a different winner.
We’re calling it BenchBench.
CLOSED retweeted
AI VIDEO HAS A PROBLEM NOBODY TALKS ABOUT.
You can generate any shot in seconds. But change one small thing in it, and the whole shot changes with it.
Swap an object, fix a detail, adjust one element, and the model regenerates everything around it. Lighting, framing, motion, all of it. So you end up re-rolling until you get lucky.
That’s the gap between a cool demo and a real workflow.
Higgsfield Miracle does the opposite:
→ Splits any video into layers, so you edit one part and leave the rest alone
→ Works at pixel level, in real time, so you see the change as you make it
→ Has world knowledge built in, so edits make sense in the scene
→ Keeps the image depth that cinema-grade post-production needs
Why it matters: post-production runs on control. Editors don’t want a new shot every time they change one thing. They want the shot they already approved, with one thing different.
It’s powered by GPT-6.2 Astra, per Higgsfield. Private beta for now, with select testers. Coming soon to Higgsfield MCP
We built Higgsfield Miracle to address one of the biggest challenges in AI video: editing a specific part of a shot without changing the rest.
Powered by GPT-6.2 Astra, it splits any video into layers for pixel-level editing in real time, with world knowledge built in. Every shot keeps the image depth that cinema-grade post-production needs.
Now in private beta with select testers. Coming soon to Higgsfield MCP.
Set up Higgsfield plugin in ChatGPT to be the first to try it.
CLOSED retweeted
this is how special effects were done before AI.
We built Higgsfield Miracle to address one of the biggest challenges in AI video: editing a specific part of a shot without changing the rest.
Powered by GPT-6.2 Astra, it splits any video into layers for pixel-level editing in real time, with world knowledge built in. Every shot keeps the image depth that cinema-grade post-production needs.
Now in private beta with select testers. Coming soon to Higgsfield MCP.
Set up Higgsfield plugin in ChatGPT to be the first to try it.
CLOSED retweeted
GPT-6.2 Astra is LIVE on Higgsfield.
OpenAI’s new frontier model for your most demanding projects.
With Higgsfield plugin, GPT-6.2 Astra powers proactive multi-agent workflows across design, coding, and creative production.
Join Higgsfield with up to 65% off for Black Friday.
Some of you have been asking very politely. Some of you have just been asking.
GPT-6.2 Astra and Sol are now available.
Super efficient and built to perform where it counts—the real world.
openai.com/index/introducing…
CLOSED retweeted
a parsing benchmark tells you which parser won. it doesn't tell you how "winning" got decided.
Reducto's BenchBench looked at that part. a leaderboard gives you one number per parser. BenchBench does the opposite:
→ 8 parsers
→ 12 public parsing benchmarks
→ every benchmark graded two ways: with its own rules, and with an LLM judge
→ then the whole thing rerun with the pages tilted 2 to 4 degrees
a small change to the test can change the result. that doesn't make a ranking useless. it makes it a starting point, not the answer.
the answer is your own documents. ParseJudge is free and open source: run your files through it and see.
Last week, I said every parser is #1 on its own benchmark. So we tested the benchmarks.
We ran 8 parsers across 12 benchmarks, then tilted the pages by 2–4°. 6 benchmarks got a different winner.
We’re calling it BenchBench.
CLOSED retweeted
EL COFUNDADOR DE CAL AI HA ENSEÑADO CÓMO HARÍA HOY SU UGC: CERO CREADORES, TODO CON IA
Cal AI llegó a 15M de descargas y a un año de 40M $ antes de que MyFitnessPal la comprara.
Llevó sus programas de UGC e influencers a la antigua, y ahora dice que este sistema le habría hecho 10 veces más productivo.
• El ciclo
> Claude Code + Apify sacan los vídeos atípicos de tu nicho, los que hacen 12 veces los números normales de una cuenta.
> Claude monta una galería local, tú eliges qué clonar, y un analizador desglosa el gancho, la tensión y el desenlace por segundos.
> Higgsfield genera el presentador con IA a partir de un fotograma de un creador real, Seedance 2.5 lo anima, y se hace una toma de prueba antes de cualquier tanda.
• Lo que lo convierte en un ejército
> Los vídeos terminados van directos a Postiz vía MCP, un borrador por creatividad y por plataforma.
> Claude escribe los textos, propone el calendario, tú lo apruebas y lo publica en TikTok, Reels, Shorts y Facebook.
> Cada plataforma devuelve los resultados: los que funcionan se rehacen con creadores reales, los que no mueren baratos.
Todo funciona en una sola ventana de Claude Code -> investigación, generación y distribución sin ningún humano en el proceso hasta la aprobación.
CLOSED retweeted
A Google Cloud engineer just showed how to build a full app with Claude from scratch.
He spent 26 minutes live on stage doing what most teams take weeks to do.
It's worth more than any $500 vibe coding course. No team, no setup, just Claude and a goal.
The people learning what Claude can actually do are shipping what everyone else outsources to a team.
CLOSED retweeted
EL CEO DE REDDIT LLEVA CASI UN AÑO SIN ESCRIBIR UNA SOLA LÍNEA DE CÓDIGO A MANO.
Steve Huffman. Programador desde niño. Primer amor: el código. No ha tocado una línea desde diciembre. Y dice que ha producido más que nunca.
"Mi primer amor es la programación. Lo he hecho casi todos los días desde que era pequeño."
"No he leído, y mucho menos escrito, ni una línea de código desde diciembre."
"Pero he producido muchísimo más."
Este investigador de Anthropic acaba de declarar que él no escribe más código, lo hace todo su Claude Code.
En esta charla GRATIS de 30 minutos te enseña a darle el contexto correcto, elegir qué partes dejarlo programar solo y a comprobar que el código sin tener que leerlo.
Esto destruye cualquier curso pago de vibe coding de $500.
Con subtítulos en inglés y español.
🔖 Guardalo cuanto antes, me lo agradecerás.
CLOSED retweeted
30 AI websites that feel "illegal" but are totally legal
1. raphael.ai — Unlimited free AI image generation, Midjourney-level quality
2. krea.ai — Real-time AI image generation, draw and generate instantly
3. magnific.ai — Unlimited AI image upscaling, details filled in automatically
4. clipdrop.co — One-click AI background removal, relighting, and object erasing
5. elevenlabs.io — AI voice cloning, mimic any voice in 5 seconds
6. suno.com — Enter lyrics and automatically generate full songs
7. runwayml.com — AI video generation pioneer, free Gen-3 trial
8. klingai.kuaishou.com — Kuaishou's Kling AI, the smoothest Chinese video generation
9. d-id.com — One photo + one audio clip = a talking digital avatar
10. sadtalker.ai — Turns still photos into talking-head videos
11. cursor.com — AI for writing code, free quota is enough for daily use
12. bolt.new — Build websites just by talking, no code
13. v0.app — Vercel's AI frontend generator, describe it and get the page
14. replit.com — Code in the browser + AI assistance + one-click deploy
15. napkin.ai — Paste text and automatically generate infographics
16. gamma.app — One-click AI presentations, say goodbye to slide-deck hell
17. notion.com/product/ai — Writing, summaries, and translations with Notion AI
18. uizard.io — Sketch a wireframe and AI turns it into a real website
19. perplexity.ai — AI search engine, answers any question instantly + sources
20. phind.com — AI search built for developers, solves code problems as you search
21. otter.ai — Real-time meeting transcription, 300 free minutes a month
22. opus.pro — AI short-video editing, automatically finds the best clips
23. poe.com — One website to use GPT-4o / Claude / Gemini
24. huggingface.co/spaces — Free playground for open-source AI models
25. remove.bg — One-click AI background removal, image ready in 1 second
26. cleanup.pictures — Erase any object from a photo with AI
27. deepswap.ai — AI face swap, video ready in 1 minute
28. musiio.com — AI music tagging and recommendations
29. jasper.ai/free — AI writing assistant, free version is enough for daily use
30. claude.ai — Anthropic's free AI assistant, the ceiling for long-text processing
CLOSED retweeted
RECAST ANY CLIP WITH YOUR OWN PEOPLE, PLACES AND PRODUCTS
up to 4 photos.
no markup, you pay what the render costs them.
studio for you, CLI/SDK for your agents.
"14.5× cheaper" is their claim. i want to see the receipts
🤝 Paid partnership
general foundation models can do pick and place really well atm.
stoked to see what a specialist model does on precise data center work like this @ninesrobotics
🤝 Paid partnership
Today, I am launching @ninesrobotics with @tianx2li.
We build autonomous robots for AI data centers.
Currently, AI only optimizes itself in software. By building the automation layer for server rooms, we are creating the world’s first fully software-defined, physically adaptive AI factory - the essential bridge that allows agentic AI to autonomously manage, manipulate, and optimize its own physical hardware stack.
CLOSED retweeted
Antes de echarle la culpa a Opus 5.5, aprende estos 15 comandos oficiales de Anthropic
15 imprescindibles para sacarle más partido a Opus 5.5:
menos tareas tediosas → agentes en paralelo → iteración más rápida
chekea esto ↓
01 /goal → sigue trabajando hasta una meta definida
code.claude.com/docs/en/goal
02 /batch → reparte cambios grandes entre agentes en paralelo
code.claude.com/docs/en/agen…
03 /subtask → delega con el contexto actual
code.claude.com/docs/en/sub-…
04 /deep-research → consigue investigación con fuentes
code.claude.com/docs/en/work…
05 /design → crea artboards editables
code.claude.com/docs/en/comm…
06 /plan → revisa el plan antes de cambiar el código
code.claude.com/docs/en/perm…
07 /memory → gestiona las instrucciones del proyecto
code.claude.com/docs/en/memo…
08 /compact → comprime el contexto en torno a las decisiones clave
code.claude.com/docs/en/cont…
09 /branch → explora una rama de conversación aparte
code.claude.com/docs/en/comm…
10 /rewind → restaura un punto de control guardado
code.claude.com/docs/en/chec…
11 /mcp → gestiona las herramientas conectadas
code.claude.com/docs/en/mcp
12 /loop → repite comprobaciones en una sesión activa
code.claude.com/docs/en/sche…
13 /schedule → automatiza tareas recurrentes en la nube
code.claude.com/docs/en/rout…
14 /simplify → limpia el código con revisores en paralelo
code.claude.com/docs/en/comm…
15 /remote-control → continúa la sesión local desde el móvil
code.claude.com/docs/en/remo…
empieza por aquí: /plan → /goal → /simplify
CLOSED retweeted
24 HOURS OF FREE UNLIMITED MINIMAX H3 ON ENVATO
starts 19 Nov, 8AM EST.
no credit cost.
all you need is a free account today.
set the reminder now, not on the 19th
🤝 Paid partnership
Coming soon on Envato. Free unlimited @MiniMax_AI H3 for 24 hours. For everyone.*
Starting from 19 Nov at 8AM EST.
Create a free account to get access. Get ready to create.
CLOSED retweeted
Instead of scrolling this weekend, give this 15 minutes.
This youtuber shows you how to build a FACELESS YouTube channel with Claude Fable 5.2, step by step.
SAVE THIS SO YOU DON'T LOSE IT
most free promos you only hear about once they're over.
Envato does the opposite:
→ free unlimited MiniMax H3 for 24 hours*
→ starts 19 Nov, 8AM EST
→ all you need today is a free account
set a reminder and get in early.
🤝 Paid partnership
Coming soon on Envato. Free unlimited @MiniMax_AI H3 for 24 hours. For everyone.*
Starting from 19 Nov at 8AM EST.
Create a free account to get access. Get ready to create.
RETELL WENT FROM $10M TO $100M RUN RATE IN 17 MONTHS
profitable.
$4.6M raised.
and customers build their own agents without depending on them.
this is what a product that solves the hard part looks like
🤝 Paid partnership
Retell just crossed $100M in revenue run rate.
$10M to $100M in 17 months. Profitable. $4.6M raised.
From day one, we’ve focused on solving the hardest parts of deploying voice AI in the product itself, so customers can build, manage, and improve their own agents without depending on us.
Our bet: the future of the contact center belongs to the teams who run it.
Proud of our team. Grateful to every customer who believed in us.
























