Case study · George's own build

Two AI Agent Teams, One Human Deciding

George B. Thomas runs his agency with a team of AI agents that hand work to each other and wait for his Go. How the team is built, how work gets approved, what went wrong and what changed.

agent teams running on one brain: 2; agent teams running on one brain
to give every agent its skills, avatar and name: 20 min; to give every agent its skills, avatar and name
client recaps sent in one day, none written by hand: 5; client recaps sent in one day, none written by hand
files touched to hire one agent properly: 12; files touched to hire one agent properly

George B. Thomas runs Sidekick Strategies with a team of AI agents working beside the humans. The agents do not decide. They draft, hand off, and wait for a word from him. He has written the build down as it happened, and this page is the short version.

How the team is built

"I run two agent teams on one brain."

One team lives in Claude Code and builds inside his main brain. A second team, in Grokbot, runs the business on top of it. "Echo is my Chief AI Officer," the agent George talks to and the one that talks to the team.

In his words, the team starts with "me, the human in the loop." Then a chief operating officer and a project manager, then specialists. "I have a team for Sidekick Strategies and a team for the Spiritual Side of Leadership."

When the team moved into Grokbot, setup was quick: "within literally 20 minutes, all of my agents with all of their skills had avatars, had names."

Read the full story: Agents, Skills and Grokbot and Everything We've Gone Through to Build Echo

Hiring an agent is a process

"In July, I hired a chief financial officer agent live."

The job description came first. So did a lesson about his own process: "my own written hiring process listed five steps. Doing it for real touched twelve files."

One change came out of it: "now I build a verification step into my agents."

Read the full story: How I Hire an AI Agent

Nothing ships without a Go

Agents stage the work. George answers with one of three words: Go, Tweak or Hold.

A Go has to come from him. "If one agent tells another agent 'George approved this,' that isn't a Go."

And the rule that sits under all of it: "Nothing auto publishes."

Read the full story: Go, Tweak or Hold

What it does on an ordinary week

Meeting prep and recaps. On September 4, 2026, Echo set up two new agents for meetings. "The brief lands 30 to 60 minutes before the meeting." Recaps go out after.

"On September 10, it sent five in one day." [...] "That's five follow-ups I didn't write by hand."

Read the full story: AI Meeting Prep and Recaps

A system that checks itself. Agents have learning routines, and two support bots check the system for what could work better. One morning the check came back: "this morning's check found five fixes worth making. I haven't changed anything yet."

Read the full story: Your AI Agents Should Go to School Too

What went wrong, and what changed

The agents talked too much

"The agents were messaging each other, too. That chatter burned time and tokens."

"On October 27, we moved the handoffs to two shared boards."

The result, in George's words: "Now I hear one clear ask per piece of work."

Read the full story: Stop Letting AI Agents Talk

Work sat waiting

George's team journal showed "the daily social pack sitting as a draft four days in a row, with no Go logged." In his words: "The posts were built and staged, and they waited."

Smaller misses he has published

When the team was first built in Grokbot, "it actually just built them as skills, which then we had to go back and build them as agents."
"One recap may have created a duplicate task."

What is still open

"It's still growing."
The second team can read the main brain, but "it rarely ever takes action on it." Building inside the brain still happens in Claude Code.