Many small AI assistants instead of one big one
Trying to give organization and memory to a multi-agent structure
I keep hitting the same wall, and I think I have been solving it backwards.
When I work with an AI assistant on anything that lasts longer than an afternoon, the conversation gets long, and somewhere in that length it stops being useful. It forgets what we settled on Tuesday. I end up being the memory. I re-explain the same context over and over, and after a while the re-explaining is most of the work.
The obvious fix is to give it a better memory. I am starting to think that is the wrong fix.
What I keep sketching instead is not one assistant with a longer attention span. It is several, each with a smaller job, and some shape between them. One that only holds the plan and never does the work. One that owns a single area and knows that area properly. One that gets a single task, does it, says what happened, and goes away.
That last one is the part I find most interesting, because it is disposable on purpose. It does not need to remember anything. It needs to be told enough.
The appeal is not that this is more powerful. It is that each piece has less to hold, so there is less for it to lose.
What I do not know yet is...