Adrien Maret

The #Skill pattern is so popular and convenient, #Google decided to integrate it directly with #Chrome I really like the Skill pattern because it's simple elegant and portable. No more vendor lock because I had to configure a tool that I use and I using the competitor would force me to adapt my existing workflow. With Skill, I can simply change from Claude Code to Codex to Chrome, just natural language and a little bit of properties. It's convenient for users but a threat for providers because churn from a product have never been easier. blog.google/products-and-p…

98.84% uptime for #Anthropic is low but can we even trust those numbers? 🙄

Am I the only one not buying the "OMG OPENAI/ANTHOPIC IS COMING FOR YOUR STARTUP" ? 🙄 OpenAI and Anthropic have a LOT of cash and they are very good at creating models and (maybe) web chatbot but they CANNOT magically replace pure players combining years of product and market experience. So yes, Anthropic just released a Vibe Coding web app like Lovable or Replit, good for them but they are extremely late on this B2C market dominated by companies who raised millions. Maybe they should have try the B2B market but of course it's more complicated to comply with Enterprise app standards than with Startup app standards (Lovable and Replit users). Not all experimentations from those companies turn into a successful product, look at Sora. Anyway, keep building everyone, don't listen to everything you hear on X.

The secret to better AI memory isn't remembering more: it's knowing what to forget. Hippo gives AI agents biologically inspired memory: decay, retrieval strengthening, consolidation. Memories that matter survive. Mistakes get amplified. Knowledge travels between tools. Reminds me of "Generative Agents" (Park et al., 2023), where AI agents lived in a Sims-like world with memory, reflection, and retrieval driving emergent social behavior. Hippo brings that same idea to real dev workflows. arxiv.org/abs/2304.03442 Open source, zero deps, SQLite-backed. github.com/kitfunso/hippo…

Best pattern I've seen recently for AI agents working with files: @mintlify built a virtual FS in RAM. The agent uses grep, cat, ls like on a real system but everything is translated into vector DB queries. Boot: 46s → 100ms. Cost: $70k/yr → ~$0. The agent doesn't need to know the FS is fake. It uses the same tools as a dev. mintlify.com/blog/how-we-bu…

@grimP1x @cursor_ai I'm mainly working on my laptop, in my IDE and so with my terminal + Claude Code is the best coding agent on the market and it should stay like this considering the amount of money Anthropic is throwing at the project

@aschen93 @cursor_ai what made you choose Claude Code over other multimodal agent CLIs?

Goodbye @cursor_ai , thanks for the ride. Bugs, slowness, crashes, it feels like using a fork of Eclipse 🤮 On the Agentic side, Claude Code is 10x better 🙄 Time to go back to good old VSCode

@pratikbodkhe @cursor_ai @pratikbodkhe 100% agree, just hope Claude Code will not end up bloated (Cursor was not at the beginning)

@aschen93 @cursor_ai cursor peaked early, then got bloated. claude code just works

Just one command and my project dir go from 32 Go to 3 Go. Get ride of the crap people find . \( -name "node_modules" -o -name ".next" -o -name ".venv" -o -name ".yarn" \) -type d -prune -exec rm -rf {} +

@tetsuoai Which benchmark are you refering ? Just a screenshot without reference means nothing 😐

After nine years of development, OpenAI released a model that underperforms compared to one of xAI's initial models, despite xAI being only two years old. Let that sink in.

Vibe Clouding is the new trend 😄

@melvynx What is you token repartition? Mostly Opus or Sonnet?

Claude : our users use Claude Code too much Also me : use 87$ in 5 hours with a 100$ subscription plan. 🤔

If you are cooking with #Claude Code #SDK, you can send images using this snippet. gist.github.com/Aschen/ffe6de3…

@OpenAI just self-XSS’d 😄 A funny reminder that security shouldn’t take a back seat, even in the middle of the AI gold rush.

@marclou Until the stock market goes down of course, if you need liquidity from your stocks around 2008 it would have been complicated

Everyone around me is buying houses right now. They tell me it’s the smartest thing to do because "real estate always goes up." But when I mention investing in the S&P 500, they say it's too risky because I could lose everything. It’s not a perfect comparison, but if you put $100K down on a $500K house, or invest the same amount in the S&P 500, after 20 years—accounting for inflation, taxes, and fees—the stock market historically gives you a much higher return. I don’t know. Maybe I missed something.

@_avichawla Thanks for the wrap! I'm wondering the effects of the quantization on the retrieval quality though

That's a wrap! If you found it insightful, reshare it with your network. Find me → @_avichawla Every day, I share tutorials and insights on DS, ML, LLMs, and RAGs.

A simple technique makes RAG ~32x memory efficient! - Perplexity uses it in its search index - Azure uses it in its search pipeline - HubSpot uses it in its AI assistant Let's understand how to use it in RAG systems (with code):

Generating frontend code is easy, but for the backend it's another story (if you care about maintainable code ofc) Task was: add a "type" enum to the Task table Backend: 35 files modified (mostly tests) Frontend: 5 files modified

I upgraded @cursor_ai (1.3.9) and now I have UI bug 🫥

Biggest problem of 2025, regressions introduced by AI coding tools 😬

@lgrammel Thanks @lgrammel ! It was working with the previous versions, that's why I was surprised.

@aschen93 #issuecomment-3149670561" target="_blank" rel="nofollow noopener">github.com/vercel/ai/issu…

⚠️ Careful when upgrading #Vercel AI SDK to 5.0.0 There is a problem with the types of the `generateObject` function that cause the Typescript Server consume 100% CPU with "Type instantiation is excessively deep and possibly infinite". Better stay on the 4.3.16 version