My current agent setup, in Slack/Google Chat:

- a channel with a team of 6 agents that acts as an infra team, managing other agent runtimes, environments, tools, skill development and services (like daily reports, google cloud, etc)
- small “salon” teams that are spun up for specific tasks, long horizon experiments, creative ideas or quick execution on a new idea, they work autonomously, and have their own schedules, they have tools to report problems to infra
- salons have different structures and numbers of agents and different mixes of models to observe dynamics (like 3 workers 1 chair, all workers with opposing priorities, etc)
- direct one to one agents for fast human coworking on specific tasks
- a voice agent that I can talk to, it can look at any of these channels, send messages, generally manage through chat
- any model/harness easily spun up

Mostly runs itself, biggest problem is context management.