Is anyone else mental like me and constantly juggling several simultaneous projects and multiple AI subscriptions? I have too many ideas so have 3-4 going at once. One is always my asset desk which is the hub for all my sounds and art assets. Then there are one to two projects I consider "in-flight" and often one in some sort of experimental stage. I utilize a Claude 20x account, a Codex 10x account and a Muse Power account.
Juggling all of this was a chore, especially as I can switch what ideas are active and what are not. I also found projects running into the same issues and making some of the same mistakes. I stupidly took several weeks to figure this out, but I eventually created myself an Overseer and a Dashboard that runs on a local webserver.
The overseer has a few important functions, though it doesn't directly control every aspect.
- There is a queue system where projects can check out an agent (Opus Medium, Sol High, Sonnet High, etc.) from the pool I have made available. Check-outs are for a fixed period of time and can get extended. The overseer can be given a priority list if I am favoring one project over another in case of contention.
- It periodically checks the context of itself and each project lead and suggests when a lead should begin the handoff process.
- Twice a day it polls each active project lead with the following questions:
- What went well today?
- What went wrong?
- What process changes, new tools, or skills would increase either your future development speed or quality?
- Record the estimate given for each agent turn and the actual time it took to complete the turn.
- Collect any decisions needed from me from any of the projects.
- Onboards and offboards projects as I active or pause them.
The data from its questions populate the dashboard. It shows me any decisions I need to make with the context behind the question. I can answer from the dashboard and the answer is routed back to the project. Similarly, twice a day I get a list of reflections - the results from the questions asked of each lead. I can Choose to queue, defer or reject each. It also identifies for me if a given process or tooling change would benefit one project or multiple. If a new project is activated it makes sure it inherits shared skills, tools and processes that have been identified as benefiting all leads.
I am probably forgetting some other things, but I have found the above very useful! If anyone else has done something similar, what things have you added that I haven't thought of (or maybe forgot to mention)?
评论 (0)