E se não puderem comparecer em nenhum desses eventos, não deixem de assistir o vídeo onde eu explico Vector Databases, Vector Similarity Search e Embedding Models de forma muito simples e intuitiva:.

Vibe coding is the WORST IDEA of 2025 according to Dave Farley:

Generate videos in just a few seconds. Try Grok Imagine, free for a limited time.

Just because an airplane doesn’t fly the same way a bird does, does it mean it doesn’t fly?.

RT @giovannibassi: Eu adorava o Kindle. Até hoje. Descobri que estou preso em mais um muro de big tech. Recentemente a Amazon nos tirou o d….

For me, I’m going back to writing handcrafted code. More intentional, built to last, and definitely more satisfying. Coding has never been the boring part of my job anyway.

I get why it works for people who don’t code though. Watching something take shape and start working is exciting, and if you’ve never had to think about scaling or long-term structure, it’s not something you’d worry about.

Another thing is optimization . or the lack of it. For a proof of concept, it can be fine. But if I want real control and need to make sure a system is solid not just from a business perspective but also technically, I have to slow things down.

After three weeks using Claude Code, I can say vibe coding just isn’t for me. My biggest problem is that it moves too fast. It gets to the result I want, but it’s hard to follow the design choices along the way. That makes it too easy to end up with messy, hard-to-maintain code.

RT @starbuxman: the new Spring AI release, 1.0.1, is packed with goodies .

RT @Redisinc: Choosing between exact and approximate search? Here’s what you need to know ⤵️. In AI applications, the way you search for si….

So yes — high context may simplify chunking. But retrieval intelligence remains critical. GPT-5 makes RAG more powerful, not obsolete. The “RAG is dead” crowd will have to wait.

That’s why tools like Redis LangCache are exciting. Semantic caching lets agents reuse answers for similar queries. Why pay twice for the same answer?.

And costs matter. Nobody building real, scalable AI apps wants to pay for 256k or 1M tokens every single request. Applied AI engineers look forward to minimizing tokens per request — and always will.

Dumping hundreds of thousands of tokens blindly is:. ⚠️ Inefficient.⚠️ Expensive.⚠️ Risky for quality (irrelevant data hurts answers). You still need smart retrieval to decide what goes in the prompt.

But here’s the catch:. Even 256k tokens isn’t infinite. Anthropic says: 1 PDF page ≈ 1.5k–3k tokens. That’s just ~83 pages. My insurance policy? 40 pages. That’s ONE doc.

Bigger context means:. ✅ Less aggressive chunking.✅ More coherent passages.✅ Fewer retrieval calls.✅ Simpler orchestration logic. It’s not the death of RAG — it’s RAG with less friction.

RAG (Retrieval Augmented Generation) = giving an LLM fresh/external data it didn’t have during training. 📂 Private docs.📅 New info.🌐 Web/API calls. If you’re enriching prompts with anything outside training data, you’re doing RAG.

High context windows might kill fancy chunking strategies. But kill RAG itself?. ❌ Impossible. In fact, GPT-5’s context size is a big win for RAG.

🚀 GPT-5 is out — and that’s AMAZING news for RAG. Every time a new model drops, people declare “RAG is dead” because of bigger context windows. This time they’re also pointing to better accuracy with massive token counts. Let’s talk about why they’re wrong. 🧵

Na quadra da casa onde moro na Holanda só tinha 6 carregadores para carros elétricos. Pedi para a prefeitura para que instalassem mais dois. Um mês depois responderam com o local onde seria instalado. Esse mês deram a data para instalação. Toda a informação é pública @sseraphini

"GPT-5 opens a whole new world of vibe coding.".