What Levelsio Wants, I Already Have

Yesterday, Pieter Levels tweeted about trying to get “vanilla Claude” to automate his entire startup creation process:

claude -p “come up with 1000 startup ideas from the top Reddit posts, build their landing page, reg domain names and add them as vhosts to Nginx on a VPS you make on Hetzner, add Stripe buy button, test in Chrome, don’t make any mistakes” –dangerously-skip-permissions –chrome

Claude, of course, politely refused. It cannot create accounts, enter payment info, or register domains. The replies were predictable: “skill issue,” “distribution is the hard part,” and my favorite, “you got a bad Claude instance.”

But here is the thing. I actually have this stack.

The Setup

I run Clawdbot , an always-on Claude agent that lives on a server and talks to me via WhatsApp. It has access to a set of skills that give it real capabilities:

• Domainr for creative domain brainstorming. It finds available short names and suggests relevant TLDs.
• Porkbun for actually buying and managing DNS. Not just checking availability, but registering domains and pointing them where they need to go.
• Hetzner for spinning up VPS instances. A single command and I have a fresh Ubuntu server.
• Ansible for deployment. My server-setup repo has playbooks for Caddy, n8n, and various apps. Run bootstrap.yml and the server is hardened and ready.

The difference between vanilla Claude and this setup is tooling. Claude can absolutely orchestrate the workflow. It just needs access to the APIs and credentials to actually execute.

What It Looks Like

When I want to spin up a new project:

1. I tell Clawdbot what I am building
2. It searches for available domains via Domainr
3. I pick one, it registers via Porkbun
4. It creates a Hetzner server
5. It runs Ansible to configure everything
6. DNS gets pointed to the new server

All through chat. No terminals, no dashboards, no context switching.

The Missing Piece

Levelsio’s prompt was ambitious but vague. “Come up with 1000 startup ideas” is exactly the kind of request that produces slop. The insight from Vibe Code Camp this week is that we forgot to plan.

The compound engineering approach is: plan, work, review, repeat. Average prompts produce average results. If you want Claude to build something real, you need to be specific about what you want and review the output at each step.

The automation is not the hard part. The taste is.

Why This Matters

We are past the point where AI coding is a toy. The infrastructure exists to give Claude real agency over your stack. The question is whether you trust it enough to hand over the keys, and whether you have built the scaffolding to make it useful.

The replies to Levelsio’s tweet missed the point. The problem is not that Claude cannot do this. The problem is that most people have not built the bridge between Claude’s capabilities and their actual infrastructure.

I have. And honestly, it is more fun than I expected.