LumiFumi

How do you imagine safeguards and some kind of guidance?

E.g. you give it some code and what does it do? does it to multiple tasks in one go or does it decide to have one task?

What I can imagine is that this only works in 2 Parts, a Agent who creates the to-dos another agent to steer it, i see a large need to beat it out of the model to see "always a gap" or "always a to-do" if i could let this run across my entire codebase with a "yup looks good" then I'd like some verification

Because already a model can be easily triggered with "here is the current list of grafana alerts" and it will know to act on a grafana alert