My brand, run with AI agents.
I am using AI agents to operate my own personal brand.
The agents handle defined tasks. I give ideas, make decisions and review what is scheduled.
I publish how the system works and what it produces.
One repository runs the operation
Everything starts in a private GitHub repository.
It contains:
- The plan.
- A to-do list.
- A timeline.
- Decisions.
- Rules for writing and choosing posts.
- Publishing schedules.
- Things the agents must avoid.
Claude agents read these files at the beginning of every session.
That gives each session the current plan and the same working rules.
How content gets published
1. I give the direction
I share ideas and notes. I decide what the brand should cover.
2. Agents prepare the content
Claude agents turn my notes into drafts and score them against the written rules.
Each draft receives points for hook, proof, pillar fit, audience fit, sales bridge and distinctiveness.
The minimum score is 11 out of 18.
3. A script checks the drafts
A small script looks for long dashes, common AI phrases and excessive hashtags.
4. Agents schedule the posts
The agents use LinkedIn's and X's native schedulers through my signed-in browser.
They prepare posts five weekdays ahead. Scheduled tasks run on weekdays.
5. I review the schedule
I can edit or delete anything that is scheduled.
I also use ChatGPT for design and a second opinion.
Day 1: The system took shape
The agents wrote the plan, the rules for choosing and scheduling posts, and the first voice guide in one day. I decided what went in.
A mistake in the voice guide
I initially asked an agent to learn my writing style from older LinkedIn posts.
Those posts had also been written with AI.
The agent was learning from previous AI output.
I switched the source to my real chat messages, where my natural wording was easier to identify.
The first numbers
| Metric | Result |
|---|---|
| LinkedIn posts | 3 published, 4 scheduled |
| X posts | 5 published |
| X views | 6 to 14 per post |
| X followers | 6 |
| Newsletter | 2 editions, 249 subscribers |
| My time | About 6 hours over 3 working days |
| Agent failures | 2 |
The first Instagram and Threads posts are published. The first TikTok post is scheduled.
Three LinkedIn articles are scheduled for 11 November.
Two failures in the first days
On 6 November, an agent stopped before an action that required human approval.
On 10 November, another agent prepared three LinkedIn articles but ran without access to the browser needed to schedule them.
I document both incidents and the changes they require.
Update history
11 November: This website went live
The site launched with its first pages, the weekly operating numbers and three experiments.
10 November: More channels and public brand rules
The first Instagram and Threads posts went live. The first TikTok post was scheduled.
I published the visual brand rules at tomakadlec.com/brand/.
The rules are public so people and AI tools can use the same source when creating brand material.
6 November: The experiment started
The agents drafted the operating plan, scheduling rules and first voice guide. I made the decisions.
I also corrected the voice-training source from old AI-assisted LinkedIn posts to real chat messages.
What comes next
Two tests are planned:
- Launch the first free tool.
- Publish the completed Week 1 report on 13 November.
I will add each result to this page.
Follow the experiment
I share what the agents do, what I decide and what the numbers show.