Before
One doer
Me + a chatbot, starting from scratch every time. Context died with each conversation. Nothing accumulated.
Not one chatbot. An org chart of them — a chief of staff that sees the whole board, and specialists it delegates to. This is how a one-person life starts to run like a company: each agent owns a domain, drafts the work, and stops at the line only a human should cross.
For a year I used AI like everyone else: one chat, one task, repeat. The shift came when I stopped asking "do this for me" and started asking "who on my team should own this?" — and then built the team. A chief of staff at the top. Specialists underneath. Each with a written-down version of my own judgment, so they can make the small calls and hand me only the ones that matter.
Me + a chatbot, starting from scratch every time. Context died with each conversation. Nothing accumulated.
An org of agents, each owning a domain, each with memory, ledgers, and a daily rhythm. Work compounds.
The win isn't speed. It's that a single person can hold six lanes of life at once — and not drop the ball.
The top of the chart sees the whole board and routes work. Each specialist owns one domain and inherits the same "virtual twin" of how I think — so nothing contradicts itself.
No agent is a clever prompt. Each is a small, durable system that the next one can copy.
Markdown files that write down who I am, how I decide, and the voice to report back in. Specialists inherit the top-level brain instead of duplicating it — one source of "how Gimbo thinks."
Each agent is a skill I summon by name. It reads its brain, does the domain work, and reports in a tight, scannable format — never a wall of text.
Open loops, a portfolio, a content pipeline — living files, not vibes. State survives between sessions, so the work compounds instead of resetting.
Scheduled jobs wake the right agent at the right time — a morning briefing on weekdays, a portfolio review tied to the tax year. The fleet has a heartbeat.
Building it taught me more about delegation — to people and to machines — than any management book.
Every specialist stops before the irreversible act — no trade is ever executed, nothing is ever auto-posted. The agent does 95% and hands me the last 5% that actually carries risk.
If a fact lives in two places, it's already wrong in one. Each domain has exactly one ledger, and everything else reads from it.
If a run fails or data didn't load, the agent says so — plainly. A confident wrong answer is worse than a flagged gap. No fabricated numbers, ever.
Anything that might go public is desensitized first — no employer, no real names, no private figures. The rule is wired into the agents, not left to memory.
Specialists share one brain for "how I think," and only add what's unique to their domain. Change my mind once, and the whole fleet updates.
A fleet you can trust runs on written-down rules, not vibes — and it gives one person the reach of a small team.