So, now that the coding agent project is successfully testing and validating on every run, and most of the technical glitches have been resolved, it’s time to think about the orchestration layer.
The bigger app.
Back to the city of happy little working agents.
Studying Paperclip was healthy here. Mostly because of how minimal it is compared to the Busy39 codebase. I like the discipline of it. The focus. The refusal to become a cathedral when a good workshop will do.
The footprint is smaller than Busy39 was.
And what’s interesting is that Paperclip sort of outsources the actual agentic work to coding MCP agents, rather than attempting to own the whole stack.
Looking at it, I think this is mostly correct.
I think Busy39 collapsed because it tried to handle the entire runtime inside one monolithic system. Compartmentalized or not, it was still a monolith, and the testing surface was enormous.
In practical terms, I’m less concerned with building another agent runtime from scratch, and more concerned with hooking an app like this up to third-party agents like Codex and Openclaw, especially because they aren’t designed to be long runners.
This app needs long-horizon solvers with features that simply don’t exist in most modern agent frameworks.
But I do like Paperclip’s multi-company model. Although, honestly, if a system like this reached any real scale, I think that structure would become incidental. In practice, you would probably repurpose it for divisions, departments, teams, or projects, then tie the layers together rather than pretending they should stay fully isolated.
Two different ambition bugs
Like Busy, Paperclip also has an ambition bug.
If you try to set up too much, the system will fail.
But for different reasons.
Busy39 failed because it tried to own too much of the runtime.
Paperclip’s weak point is the manual setup. It assumes a certain amount of orchestration already exists around it.
It is not polymorphic or self-scaling in the same way an app like Busy39, or something like Hermes Agent, could be.
And I recognize that mindset.
I understand why it exists.
Understanding this was legitimately helpful.
Because going into Busy41, the shape of the problem feels clearer now:
Don’t own the whole stack.
Own the continuity.