Case study · George's own build

A Ministry Site Built and Run With a Team of AI Agents

George B. Thomas built The Spiritual Side of Leadership over a four-day weekend, then grew it with eleven AI agents and one human making the calls. What got built, what broke and what changed.

AI agents, with one human making the calls: 11; AI agents, with one human making the calls
framework pages carrying imagery by July 21: 32; framework pages carrying imagery by July 21
defects found by four AI models, all fixed the same day: 9; defects found by four AI models, all fixed the same day
mornings a week the Daily Drip sends on its own: 5; mornings a week the Daily Drip sends on its own

George B. Thomas built The Spiritual Side of Leadership, his faith and leadership ministry, with AI. He has written the whole build down as it happened, in an eight-part series. This page is the short version: what got built, how the team works, what went wrong and what changed because of it.

Where it started

The first version came together over one long weekend.

"The family was away over the four-day Thanksgiving weekend." [...] "I literally taught myself how to vibe code, dude."

That weekend produced a member dashboard, daily Scripture, soul and prayer journals, a daily presence check, a decision discernment canvas, a faith-based leadership coach and an assessment built on four cornerstones and ten pillars. When a referral system was needed, George's note on it was short: "I was like, that was seven minutes."

"By January, I was building the next layer from the terminal." "In under a minute, the article had a fillable worksheet," and by February that had become a button in the admin.

Read the full story: How It Started.

The team: eleven agents and one human

By April the build had a team.

"Eleven AI agents, each named for someone in Scripture, and one human."

George calls it "a council of twelve: me plus eleven agents." Four are Counselors (George, Solomon, Barnabas and Nehemiah) and eight are Artisans, each with a lane. The lanes include review by a human where it matters: "Deborah drafts; attorney reviews before publication."

The human stays in the decision seat. On the rebuild in July, George put it this way:

"AI agents drafted the code, the menus and the image prompts. I made the calls."

Read the full story: Twelve Biblical Names, One Human.

Rebuilding the site around the real problem

In July the site was rebuilt around the 18 leadership gaps, "the 18 places where what the business world teaches about leadership and what Scripture teaches pull apart."

July 16: the gaps went live with a new design system. The mobile menu did not show them at all. "So we fixed it twice in one day."
The same day: "the articles went free. Every one of them."
By July 21: "all 32 pillar, cornerstone and gap pages carried biblical imagery."

Read the full story: The 18 Leadership Gaps.

What went wrong, and what changed

George publishes the misses too. Three of them shaped how the system runs today.

A fabricated verse reached production

"A fabricated John 3:16 shipped to production on my ministry site."

The fix was a rule and a test. By late September the rule was written into how the team drafts every article: Scripture is retrieved, never generated. On October 10 the team went looking for every place a model could still write a verse, and added a regression test: "If any file asks a model for verse text, the test fails." On October 12 George made the call that closed it out: "WEB everywhere, everything public."

Read the full story: No AI Ever Writes Scripture.

Four AI models were asked to attack the site

On October 11 George handed Claude, ChatGPT, Grok and Gemini the same brief. The result:

"Nine confirmed defects went to the top of the list, and all nine were fixed that same day."

George is plain about the limits of that claim: "Two of them can't be fully proven until a real charge runs through."

Read the full story: Four AI Models Attacked My Ministry Site.

An automatic email went silent

The Daily Drip "goes out on its own at 9 AM Eastern, Monday through Friday." A review email lands in George's inbox a weekday or two ahead of each one, with buttons to hold, skip or undo.

Then it missed a day.

"On Tuesday, October 20, the Daily Drip sent nothing... Nobody knew for three days."

The change: "a check runs an hour after send time."

Read the full story: The Daily Drip: Automatic, Never Unattended.

What zero taught him

On October 14 the team shipped a daily sitting. "Eight days later, it had zero sittings." George's read:

"zero wasn't a code problem, it was a reach problem."

Read the full story: A Production Studio, and What Zero Taught Me.

What is still open

George names what he has not decided or finished:

Whether silence should count as approval on the Daily Drip: "a decision only I can make, and I haven't made it yet."
"Some admin-only pages still carry typed verses, and they're on the list as named debt."
The pocket companion app is "a concept and a clickable prototype right now, not a launch."