Marcus Biel

@marcusbiel.bsky.social

Rockstar Dev turned into SaaS founder, working on Email Marketing for small businesses. Tweets: email marketing, startups, coffee & survival as a dad, ADHD, 🇺🇦 https://maildroppa.com

Building Maildroppa with GPT Codex + Claude Code. Claude is fast; GPT Pro thinks deep; they review each other. In weeks: 35-language React app, revisions, health scoring, Rspamd spam check, backend refactor for A/B testing. Heaven + hell: speed is addictive, pressure is brutal.

🚀 Maildroppa update is live After 4 months of modernization focused on stability & clarity, we just shipped: 🗣️ UI in 35 languages 🧩 Central Footer Builder 🌍 Subscriber-facing text language 🔁 Undo / redo in automations Built on a fully modernized codebase. Feedback welcome 💬

AI won’t kill SaaS—it’ll kill trivial “A→B” utilities. Real software is years of decisions, edge cases, and domain knowledge. AI can write code, but you must specify. Without deep understanding, prompts produce polished junk.

AI won’t kill SaaS—it’ll kill trivial “A→B” utilities. Real software is years of decisions, edge cases, and domain knowledge. AI can write code, but you must specify. Without deep understanding, prompts produce polished junk.

🚀 Maildroppa update is live After 4 months of modernization focused on stability & clarity, we just shipped: 🗣️ UI in 35 languages 🧩 Central Footer Builder 🌍 Subscriber-facing text language 🔁 Undo / redo in automations Built on a fully modernized codebase. Feedback welcome 💬

During a React Flow rebuild, edges went wild. Screenshots + AI (ChatGPT 5 Pro, Claude, Juni Ultimate) didn’t help. I wrote: "node top-left, another to the right; put a + at their midpoint." Instant fix. Lesson: images can mislead—precise text wins.

really miss GPT 4.5 for writing and o3 for coding. It's weird how Sam Altmann made this big statements of how super smart GPT 5 was -when it's not. Super slow is all that it is. Sonet 4.5 got quite nice though! What are your experiences?

2-week verdict on Claude Max: gorgeous code 😍, zero context 🤪. It invents APIs and gushes that you’re brilliant even when you’re wrong. GPT-O3’s code may be less pretty, but it has much more context and just works—and it’s far less of a yes-bot. Back to GPT 🤠

I get many calls on my website to "Y29uZmlybS" which is base64 encoded for "confirm". Does anyone else see such behavior? Why? Who? What to do against it? It screws my Plausible stats :(