I've always been a fan of popular science channels. And I'm donating to some. But I didn't expect the ones from the US will ever find themselves in a situation like this, too. youtu.be/LQjFGx9mEPw?si…

A new rationale for being polite in AI chats has been found! Some people claim that the results are better. Could it be right? Well, LLMs' wisdom has formed from human texts, so it is probable that good people write better overall.

Don't stick to AI too hard when using machines to generate code for you. Remember the good old algorithmic templates. E.g. Live Templates in JetBrains or scaffolding in frameworks. For the repeatable parts, it is much faster and more reliable. #devtalks

I call some of my AI flows 'agents,' which is not quite correct. Why do I? It's because when it is started, I vaguely know how to utilize an LLM right for the problem, so I give it some freedom to decide what to do, and that's a valid AI agent example at that point. But as I

A single command to create CLAUDE.md with Claude Code.

I've been crabby about CrewAI's hierarchical mode, because it is essentially non-useful. LangChain has released a library that seemingly addresses exactly what I've been missing in CrewAI. youtu.be/JeyDrn1dSUQ?si…

- Do what cafe, excuse me? Do you like ALLCAPS the same as Bitef Art Cafe does?

It is wrong to expect that AI helpers will let you work less. They will enable you to do more, so you are going to work even harder. To work less is a skill of its own.

Windsurf sandboxing is... flexible.

Windsurf has included in a scaffolded view two fields that were in the 'future iterations' section. Now I'm looking at it in Storybook and thinking 'why not...' We will clearly work more with AI-helpers, not less 😮‍💨

I'm giving a Master-Class in Belgrade 🇷🇸, this Saturday 15th. "AI Agents Write Your Code" For those who are reluctant to adapt to Cursor/Windsurf. linkedin.com/feed/update/ur…

Again and again, the boldest mistake in the evaluation of coding AI agents (Windsurf, Claude Code) is that people try them by creating an app from scratch. Using AI coders in mature projects is an entirely different experience and skill set.

One of the key powers of AI-agents that I leverage is exceptionally cheap PoC implementations.

AI is like a highly-aware human senior developer with a very-very specific form of amnesia. Join the master-class tomorrow in Belgrade. linkedin.com/posts/serbian-…

AI-agent coders unlock some dreams from the past for us. What is a blocker for perfect decomposition? Mocking the dependency to test things in isolation used to be booooooring! With Windsurf it is no longer boring, as writing the long trivial code is its evident strength.

Maintain your process checklists in a form suitable for both artificial and evolutionary types of intelligent substance.

In the past: 20% code 80% debug Today: 2% code with Windsurf/Cursor 8% review and debug 90% debug a poorly documented peculiarity of Google Cloud (or another 3rd party) that neither you nor LLM can solve analytically. 🤕 NB: the numbers are grotesque.

One of the ideas that shaped my experience was top-down design by Dijkstra. This concept is unfortunately not very mainstream, but it was critical to my skill development. And now, AI is changing this approach. Although the fundamental philosophy remains untouched, here's the

When you've fixed something that an AI-coder hasn't been able to fix. Or when you stop it before the last summary because you see its test command has succeeded. It is stumbling not to let 'the guy' know the resolution. Unfortunately, the guy is stateless, so he does not care.