latchattack

I generally try to get a few parallel workstreams going, and try to cap concurrent tasks based on effort of build/review. For me the bigger bottleneck is when a number of agents finish simultaneously and all require big reviews requiring thoughtful project context.

So i try to stagger the big rocks and chip off smaller ones in parallel. Also being able to complete the smaller tasks and move on makes me feel less brain mush from parallelizing.

I use a project decision layer too to provide guidance across sessions. This makes me feel more confident in their work staying aligned with project direction.

Wonder how others determine when agents are running normally or when they need some human judgement.

how are policies versioned? after a block is caused, does the audit leave some trace of the policy state or how it happened? i.e. are blocks reproducible

I wonder if the annotations are saved after completion/plan approval? Is the annotation helpful for single plans only or does it influence later plans

I agree that similarity search can't surface reliable project judgment alone. How do you know what was true, what was rejected, and what's obsolete since changes in the project? I am actually building Open-Latch for this reason. It leverages project decisions, facts, lifecycle state, reconciliation, enforcing cited evidence and a judgement gate before coding agents act.

GitHub: https://github.com/open-latch/latch

I felt the same annoyance as OP mentioned, but focused on agents violating project judgement and direction. Obviously we are optimizing for some different things than deja: deja uses no-LLMs and exact verbatim transcript retrieval.

Latch uses LLMs, ranking, etc, to emphasize strong judgment, correction, and enforcement over dynamic project flow. Give latch a try if you're also running into decision amnesia in your workflow.