RT @jshchnz: Who uses Claude Code on the Mac and wants to try a crazy new app?

Agree, AI agebts areil annoying with all the rules files. In my case it is better to have small separated files with a few specific rules and use one of these for each task. One for testing, other for PR check, other for architecture, etc. bye bye rules per agent.

RT @KevinNaughtonJr: i still think about this almost daily

We built something cool. My daughter had fun. But I didn’t learn much, and that scares me a little. She’s happy. The future is uncertain, devs ignoring how to code and creating spaghetti code super fast 😬.

As a dev, I didn’t grow. I became better at *using* AI, not better at *coding*. You still need solid knowledge in architecture, patterns, testing. AI won’t teach that. ➡️ And finally….

✅ What AI did well:. Fast prototyping. Good for idea discussion. Helpful on *small*, focused tasks.❌ What failed:.– No big picture.– Weak on rules.– Not a good teacher.➡️ My professional takeaway….

Small prompts = random results. If I wasn’t clear or missed details, the output was chaotic or useless. I had to stop and restart multiple times to fix things. ➡️ Let’s sum up….

The AI wants to make you happy. It won’t challenge your ideas or suggest better ones. It picks the easiest path to *satisfy* the prompt, not the *best* solution. Remember Homer Simpson asking the monkey’s paw for wishes? It feels like that 🙈🧞‍♂️. ➡️ That leads to another issue…

The more rules I added (TDD, SOLID, clean architecture. ),.the more confused the AI got. It only followed the bare minimum needed to fulfill the prompt. ➡️ And that’s not all….

The more tests, the worse it behaved. Eventually, I gave up on unit tests and asked for integration tests instead. ➡️ And now, something unexpected happened….

I enforced strict TDD: write tests *before* code. It worked well at first. The AI created good scenarios. But later, it ignored failing tests or skipped creating them entirely. ➡️ Why? Context limits and “boredom” 😅.

Tip: Ask for a refactor *plan* first. Break it down into small steps. Even then, large parts must be split further. AI can't refactor large contexts reliably. ➡️ But what about test-driven development?.

The initial code was badly structured. I asked to refactor. Good theory, bad execution. Context limits caused partial results. ➡️ Want a clean refactor? Read this 👇.

Claude created a working prototype *really fast*. But instead of using a chess library, it hardcoded the logic. Only switched to Stockfish after I explicitly asked. ➡️ Then came the architecture mess 🧱.

We used Claude Code with Claude 4 Opus. It took just a couple of days. She was excited to build something. I was curious to test AI's limits. ➡️ Let’s start with the pros ⚡.

During my holidays I played chess with my daughter. She said: “What if captured pieces *join your team* instead of disappearing?”.We tested it, loved it, and built a version together 👉 ➡️ From a developer perspective, here is how we made it with AI 👇.

RT @chess24com: Magnus: "We've got to find a new chess variant. This game has become so dominated by preparation, so monotonous — you see,….

IA entrenada solamente con temas de la biblia católica #catequesis #Biblia.

RT @umesh_ai: Prompt on ChatGPT 4o : A digital illustration of a [SUBJECT], portrayed with a network of glowing clean pristine blue lines o….

RT @FaunaNatu: Ya no hacen música como esta