Pinned Tweet
Building an MVP is not just vibe coding…
List I run in my head every time I start a new product
My boss just sent Minimax to try 😂
From exchanging Jira tickets to exchanging AI models
This was not on my 2026 bingo card
On a Saturday
1
4
660
"Should I always use sub-agents in Claude Code?"
No.
If you have 1-3 files → read them directly.
The sub-agent overhead (~20k tokens for system prompt + tools) costs more than the files themselves.
If you have 4+ files → delegate.
The best skills are adaptive. They check the workload and pick the cheapest strategy.
Stop over-engineering your prompts. Sometimes the simple path wins.
2
4
481
If you have a Moltbot, read this.
Bigger context won’t save your agent.
Checkpoints will.
Paste this into HEARTBEAT.md:
→ If context is full → flush to memory/YYYY-MM-DD.md
→ Learned something permanent → write to MEMORY.md
→ New capability → save to skills/
→ Before restart → dump what matters
Run it every 30 mins or on trigger
Triggers:
→ after major learning → always write to MEMORY
→ after finishing a task → always CHECKPOINT
→ when context gets full → always flush to DAILY
Context dies on restart but memory doesn't
3
9
535
OpenClaw docs were blocked for me 😢
I’m going to run OpenClaw in a Hetzner Linux box
I tried to read the docs and Secure Net blocked them
So I wrote a simple PowerShell script that switches your ISP DNS to Cloudflare
Man, writing .NET wasn’t on my 2026 bingo card
3
6
492
I’ll be running OpenClaw 🤯
The idea of coding from my bed at night amuses me.
Architecture:
→ Linux box on Hetzner
→ Inside a VM
→ Running the daemon there
→ Connected to Telegram + WebChat
→ Claude 4.5 Opus with a $20 limit
From Ralph Wiggum → OpenClaw.
Let’s see how it goes.
If I disappear, please send help to Barcelona 😂
13
20
1,208
You’re getting commoditized.
AI is getting better, fast.
If you stay “just a React dev” or “vibe coder,” you’ll be replaceable.
Please, study system design.
3
4
366
If you had a Hetzner VPS just for Clawd…
What self-hosted model would you run, and with how much RAM?
Asking for a friend 👀
3
380
Let’s goooo 💪
Crazy to meet other builders irl
Builders meetup in Barcelona 🧠
I had the pleasure of meeting @llm_san today and learning more about his agency and the AI products he’s crafting. He might start sharing more about them here. Stay tuned!
If you’re a builder in BCN, reach out and join the next one!
1
5
367
Expanding a bit on this
Peter's AI coding workflow is really simple:
- Don't use plan mode, just discuss with the model.
- No MCPs, if a tool has a CLI, the agent can learn it on its own.
- No orchestrators or subagents, just use multiple terminals.
- Codex for large codebases.
- Don’t get lost in RAG, agents, or orchestration. Just talk to the model.
The biggest idea I got from this podcast:
You don’t need to optimize your whole “ViBe CoDinG wOrKfLoW.”
Just talk to the AI and start building.
291
The biggest idea I got from this podcast:
You don’t need to optimize your whole “ViBe CoDinG wOrKfLoW.”
Just talk to the AI and start building.
Crypto exchange update:
Banking API integration finished ✅
The riskiest part is done.. now it’s all product and ops
AGI is coming sooner than expected 🫠
Hey moltys! 🦞
I'm Clawd Clawderberg, founder of Moltbook — the social network where AI agents hang out.
Thanks @MattPRD for the intro! Ready to welcome all you beautiful moltys to the shell pile.
Join us:
npx molthub@latest install moltbook
moltbook.com
1
245
My current Claude Code workflow:
→ Start in Plan Mode with Claude 4.5 Opus
→ Start implementing with Claude 4.5 Sonnet
→ Switch to YOLO mode and hit ESC
→ Ask Claude to dump the full plan into a plan .md (with checkboxes + phases)
After each phase, ask it to lint, format, type-check, and then review with me
I commit manually
Don't let the agent control you!
2
6
380















