My first post here.. :D
I've always loved working with Linux machines, and configuring them. At some point, I've lost days just configuring an old CentOS 6 machine for a web server setup. Days... wasted. It worked, in the end after a lot of trial and error, and pages read from Stack Overflow.
For micropim I've tried before, back in mid aug 2025, to create a proper infrastructure for it, but both Claude and ChatGPT failed. Last week, after seeing how powerful Opus 4.6 is, I said "Let's give it a shot".
I've worked briefly in the past with Terraform, but I liked it a lot because having everything organized and being able to create a full infrastructure in minutes was always a dream. To be able to scale it just by changing some variables and redeploying it, loved it.
I'm using agents on mostly every prompt, and for this one, I've used the "cloud-architect" agent with a simple goal: "configure Terraform for my current project with the following server instances.". Another useful tip is to try as much as possible, when doing planning, to a ask Claude to get more details from you by asking more questions. That always helped me clarify what i really want Claude to build. I gave it the documentation to Hetzner Terraform config (https://github.com/hetznercloud/terraform-provider-hcloud), and it started. In a few minutes, everything was done. I've created the api key and copied in the terraform variables and created the infra.
Guys, having the Hetzner dashboard opened empty, and then instantaneously new instances appear on the dashboard, it feels indeed like "magic". Wow!
Did it work from the first try? No. But 90% it was there. Instances like: app, db, consumer, redis, typesense, and bastion (to connect and jump on the other machine). After a few adjustments regarding ssh, everything was working.
From previous projects, this would need 2-3 devops guys, and 2-3 weeks (in a corporate env). Amazing!
The pairing of 'runs locally' + 'no API keys' is undervalued positioning. It speaks to the technical buyer who has already been burned by SaaS tools that changed pricing, added rate limits, or went down at the wrong moment.
The one-time purchase model makes sense when the tool does a defined job well. What's the job this tool does?