Post

I've just released github.com/freecomputin..., an extension set for your coding agent to work effectively with jj and beads.

github.comGitHub - freecomputinglab/soroban: Work with beads and jj effectively in a coding agentWork with beads and jj effectively in a coding agent - freecomputinglab/soroban

Lachlan Kermode

@ohrg.org

6 Aug

I recently described this process at length in a blog post: www.ohrg.org/taming-jagge...

ohrg.orgTaming jagged intelligence with beads and jj

Lachlan Kermode

@ohrg.org

The basic idea is that LLM agents do much better when you parcellate work for them in advance. This kind of planning also opens up more economical downstream agential workflows such as using a local model to implement work that a more capable model from one of the frontier labs has planned.

6/8/26, 9:41 am0 reposts2 likes

Lachlan Kermode

@ohrg.org

6 Aug

In addition to the process laid out in the blogpost, I've also started using jj workspaces as a way to isolate agents' work from each other while they are working on a bead. If you start using soroban, you'll get this workflow by default.