Everything I recommend works for me first. Otherwise I wouldn't be talking about it.
This isn't an assistant or a chat. It's a team: each agent has its role, and I hold the decisions and the responsibility.
And that's not the ceiling: ad-hoc subagents spin up for a specific task — researchers, auditors, builders. The team assembles around the work, not the other way around.
Open hypothesis-projects. The status is honest: some are alive, some are paused — a lab isn't obliged to ship everything.
Repeatable blueprints an agent reads and builds for you — that's idea 04, "Blueprints," in action. Not a one-off prompt but an assembly line: take a ready blueprint and run it yourself.
Your wiki, built and kept current by an agent: notes link themselves, and knowledge doesn't get lost in chats.
Several agents argue and converge on a decision — a council instead of a single assistant. The roles, the rules, and how to launch it.
Write in Obsidian — it lands on your site. The vault as the source, git as the pipe, no manual layout or copy-paste.
HQ on the strongest model, workers priced to the task, Codex as a second track: how one person runs dozens of AI sessions a day through files instead of chats.
How to talk to an LLM: context, the four levels, your own context file, and ten prompt patterns. 39 pages — from zero to a working system in one evening.