@bcherny I created a custom configuration with Claude Code on Telegram a month ago. It works incredibly well!! I even added a SOUL.md file + MCPs + Skills. All of this is free thanks to my Claude subscription.
@trq212 I created a custom configuration with Claude Code on Telegram a month ago. It works incredibly well!! I even added a SOUL.md file + MCPs + Skills. All of this is free thanks to my Claude subscription.
I replaced OpenClaw with a 200-line Python script.
Same VPS. Same Telegram. But now it runs Claude Opus with:
Voice messages (Whisper transcription)
Photo analysis (Claude Vision)
Infra monitoring (alerts me if my servers go down)
A SOUL.md file that evolves its own personality
Daily briefing at 9am
No Docker. No framework. Just a Python bot + Claude Code CLI.
OpenClaw is cool, but I didn't need 90% of it. Sometimes the best agent is the one you build yourself. @claudeai
What I actually gained:
Next.js, Claude AI, Stripe, Vercel, Drizzle, PWA
Shipping a full product from zero to prod, alone
The humility to know when to fold
Most AI startups are building features, not products.
A feature is "AI that writes emails."
A product is "AI that handles your entire inbox workflow."
The winners in 2026 will be the ones who thought product-first.
@bloggersarvesh The real edge isn't content generation, it's speed of iteration. Ship 50 city pages while competitors debate "AI content quality" and you own those SERPs for months. How many pages before you saw ranking signals?
@twicewest94 Le plus sous-estimé c'est que Claude Code décide lui-meme l'archi des sub-agents. J'ai buildé un scraper complet avec, meme pattern : tu décris le résultat, il découpe tout seul. @twicewest94 t'as géré le rate limiting comment sur 500 boites ?
I'm managing a product that would normally require 5 people to build and 3 more to run.
I'm doing it alone.
Not because I'm exceptional.
Because the tools changed.
@claudeai Best announcement !! Been getting "prompt too long" errors multiple times a day for weeks, almost rushed to tier 4 just for that. Loaded my entire project codebase in one session to test it. No degradation I could notice. At what point does coherence actually start dropping ?
@claudeai Been running parallel Claude Code sessions daily for the past 3 months. This just made my day. Any plans for dedicated tiers for builders who use Claude Code as their primary dev environment ?
@bcherny@michalbrojak@trq212 Running 3 parallel Claude sessions right now, one deploying infra, one writing tests, one handling support logic. @bcherny's 10-15 method makes sense once you stop treating it like a chatbot and start treating it like a team. What's the max before it becomes counterproductive?