go podcast()

In this episode, we dive into the dangerous "pre-launch purgatory"—that final stretch after reaching V1 but before the first paying customers arrive. It’s a period where the temptation to start over is at its peak, armed with all the lessons learned during the build. We discuss how to resist the urge to refactor your SaaS into oblivion and why shipping "imperfect" code is the only way to get the feedback you actually need.
In the second half, the conversation shifts to the challenges of maintainership. My co-host shares the hurdles he’s currently facing with his open-source project, Andurel. When you’re building in a vacuum without a clear signal from users, how do you decide which features matter? We explore the shared struggle of finding a "North Star" when the feedback loop is quiet and the roadmap feels uncertain.

This week we talk about what's new with what we're working on. And as always we cover / comment what we've found intreesting or disturbing in the last week or so.

We talk to the author of Uncloud, a tool that helps with self-hosting and managing your own infrastructure / make it easy to deploy your services to your servers. 
Links:
Uncloud on GitHub

Ramesh contacted me regarding what we've been saying lately in the pod regarding using LLM and some bad experiences we've had and maybe even some negativity etc. He wanted to give his perspective and experiences using LLMs, where it's working well for him and his team and give some tips regarding potential miss-use and what have been working good for him.

We hop into the call and start recording, and what we found, we had both issues / concerns about quality of LLM produce code. Morten is reviewing some aspect of his project before releasing the public version and found some interesting thing that would make it hard to justify leaving them there. I had very similar issues, entering into a full refactor of a Go backend server I let the LLM cook for a rare time in Go, and decided at some point that enough is enough and decided to refactor the code.
It's not like it's big surprise, and I think we're a lot in that situation. When you truly start to review the code that is generated, let's just say that sometimes it's not the best work you'd have done yourself. It's quicker though, no question there. But at what cost.
I also finally ditched nginx and installed Caddy for my production servers.