16k commits and almost ~1M lines of code produced with AI - why didn't the project become an unmaintainable spaghetti slop?
I tried to understand that and one of the main reasons has to be related to architecture, so I took a snapshot of OpenClaw every 1000 commits and analyzed the state of architecture at each "snapshot."
Here is how it looked like at 12k commits (the final stage at 15k commits looks +- same; also btw while I was writing the post there are already 16k commits in repo):
Circles are modules (folders). Arrows are dependencies (there are too many arrows but I thought they look shiny so I didn't hide them).
The main pivot in terms of architecture seems to be Jan 11 where the creator of OpenClaw, Peter, introduced the plugin architecture:
And on a podcast "Pragmatic Engineer" around ~41:40 Peter mentioned this decision: "...I really wanted to get this PR in ... it was like 15k line change - I moved everything to a plugin architecture... and I care a lot about the structure."
Here it is that PR btw https://github.com/openclaw/openclaw/pull/661
And later there were more and more plugins appearing.
Great architectural intuition, and I think this was one of the decisions that allowed OpenClaw to maintain dev. velocity considering that it's vibe coded.
P.S. just started a newsletter - first email literally asks what you want to hear about → https://vladbrok.beehiiv.com
Top comments (0)