Martin Michalek

@machal.bsky.social

Web Performance at PageSpeed.ONE FrontKon Conference. Prague based. 🇨🇿

Don't let AI start anything. Bring it in at phase two – revision, variants, discussion, or just the work itself. It's a reliable way to keep your own expertise sharp. And to hold on to the fun part of the job. I remind myself of this constantly. Do you?

I realize again and again how crucial it is to log a task description whenever I use AI in development. And at the same time, how incredibly useful AI itself is for creating that task description.

#Chrome has been really slow lately. So I went to try #Safari again. I've only been using it for a day so I can't judge. But! I couldn't even last a day with my previous attempts. It seems to me that Apple has put a lot of effort into this browser in last year...?

One of the biggest pitfalls of developing with LLM is bulk editing. It's temptingly simple, it edits beautifully contextually, including texts. But it often breaks something else and it's hard to control on a larger codebase.

Is AI a terrible copywriter? I’m rebuilding @pagespeedone.bsky.social website with Cursor & agent skills and rating what AI can do: Frontend dev ★★★ Translator ★★★ UI design ★★☆ Copywriting ★☆☆ It produces “AI copy”. I keep rewriting. Is it the tools, or my prompting? What works for you?

My current dev setup (mostly UI): — #Cursor, Agent mode. Haven't seen the IDE since v3. — 2–3 agents working simultaneously on one task. — Dictating prompts via Whisper Flow. What’s next? I’m starting to take this setup outside of dev. I need to figure out orchestration. Reading tips?