How I run Pomegra.io solo with a fully automated Claude loop SDLC
12 git repos, k8s infra, helm, gitops, dedicated server. More than 800 daily users and traffic grows quickly.
- 0 developers
- 0 QA
- 0 devops
- 0 product designers
- 0 SEO specialists
People usually assume there is a small team behind it, or at least one freelancer somewhere. There isn't. So I want to share how the whole thing actually runs, because the interesting part is not that it works, it's how boring it became.
The autonomous SDLC
It's fully automated with Claude loops. Development to production monitoring.
Develop and push to dev server? - claude loop.
Github actions tracking across all 12 repos? - same.
Github hasn't enough resources for build? - same claude loop, builds locally and pushes to the right place.
Tests on dev server? - same loop, same flow.
Changes after tests? - you know the answer =)
Push to master, check everything deployed? - guess.
Nothing here is a single clever prompt. Except the task itself, and that one is not clever at all. A failed build isn't a notification I read at 11pm, it's input for the loop.
What actually changed for me
I stopped running the pipeline and started deciding what the pipeline should do.
12 repos with a normal setup means coordination work. K8s and gitops usually mean a dedicated ops person, or at least someone on call. For me it means loops I designed once and correct when they drift.
Most of my week goes to product direction, positioning and content. The mechanical middle of shipping software, the part that eats whole teams, mostly happens without me watching it.
Where I still get stuck?
Shipping stopped being my problem. Deciding what to ship became the only problem.
I repositioned the product twice already. Both times for real reasons: no clear moat, a lot of free users and only a few of them ever pulling out a credit card. I'm still not fully sure about the current direction. There is no 100% test suite for "is this the right product for these 800 people" - of course at this point many product guys will catch me 😊
And for me automation makes it heavier, not lighter. When execution is cheap, you can chase a wrong direction very fast and very completely. 12 repos will happily follow me into a bad idea in two weeks, fully deployed, monitored, with clean CI.
So the only thing left on my plate is the thing I can't delegate (for now), and it's also the thing I'm worst at being confident about.
What I want to deliver with this post?
We can delegate to AI much more 😊 Our trust issues are the main bottleneck.
More tech details - soon
Top comments (0)