Mika Gilles @HowToPsy

How to learn real coding? If I want to build something meaningful, something that actually ships and scales, I need to learn how to code properly.

Yes, I work from home. And here’s why your employees should too. At DRIP, you can work from anywhere. Here’s the kicker: Remote isn’t just a “perk.” It’s a talent strategy. Why limit yourself to a 30-mile hiring radius when the best person for the job might be across the world

This is the stack I’m betting my AI Assistant on. Tear it apart before my production does.

Convex or Supabase for my project?

What's your Tech Stack?

I thought I could build my Behavioral AI Assistant with no-code. I was wrong. If I want this to scale, I need to truly learn coding. Not just my rough HTML, CSS, JS. To the engineers here: What’s the ONE thing you wish you’d learned earlier?

Hit my Cursor Pro Plus limit after only 2 weeks of heavy usage. Will switch to the $100 Claude code because I can't stand that and longer. I'm not technical, but that will change now, to really get my behavior #AI assistant going. … I finally need to learn coding. Any tips?

I'm curious, do you think we're in an AI bubble? We all are building something with AI to use AI to solve problems.

The Behavioral AI Assistant that I'm building is basically a lie detector for websites. Pretty wild to think about.

Hot take: Cursor feels less like coding, more like pair-programming with an overcaffeinated senior dev.

“Move fast and break things” works, if you’re ready to revert fast and often.

Today I learned the obvious: → Build logs are gold. I wasted days tweaking scraping & extraction logic blindly. The moment I actually read the logs… 80% of the issues made sense.

work-life balance in your 20s → guaranteed mediocrity

Building an AI Assistant feels like therapy: most progress comes from scrapping what doesn’t work.

I just killed my old 2-tier scraping setup (Cheerio + ScrapingFish). Looked good in theory, failed on 70% of pages. Switched to @browserless + Playwright with CDP, auto-scroll & DOM snapshots. Yeah, lost some time on that one but hope this is a better solution.

I watched my scraping pipeline turn chaos into clarity, no #LLMs needed. Just cold logic, smart filtering. Why does this matter for you? Because #AI #founder tells you to “just throw it in a model.” But: If your input’s garbage, your insights will be too, and you’ll pay for it.

Day 8 of building my #AI assistant OpenAI just gave us a glimpse of #GPT5. And if you’re only seeing “just another upgrade,” you’re missing it. This is not another chatbot. Startet this as a solo project, thinking my challenge was building the AI. Turns out, its keeping up.

Day 7 of building my behavioral #AI assistant The button said “love-section.” The reviews were hidden in a ghost tab. The CTA was nested in a <span> inside chaos. HTML lies. So today I built a DOM interpreter. It doesn’t care what things are called.

Day 6 of building a behavioral psychology #AI assistant Built a debug tool to see what my scraper actually extracts. HTML in → structure out (kinda). Tier A or B? Page type? CTA found? Now I can watch it breathe. No structure = no insight.

Day 5 of building the next behavioral psychology #AI assistant Scraping Pipeline: → Tier A: fetch + Cheerio → Tier B: JS rendering via ScrapingFish But raw HTML nukes the #GPT context window. Now building a rule engine to extract only what matters, before the model thinks.