I'm Building 8 AI Agents in Public. Here's the Plan.

On my homepage there's a dashboard showing eight AI agents.

Right now, none of them actually work.

That's not a confession buried in a footer — it's the entire premise of this site. Over the next six weeks I'm going to wire up every one of those agents to do real work on my content. Drafting, editing, scoring, scheduling. And I'm going to do every step in public.

You'll see the prompts. You'll see the schemas. You'll see what fails. You'll watch me approve a draft, reject one, and explain why.

By the end of June, the dashboard you see today will be a real working system. Until then, every badge on it is honest: wiring up, with a target date and a real progress number underneath.

Why I'm doing it this way.

Most builders ship the demo and hide the build. The polished video, the screenshot of the working product, the customer testimonial. You never see the seventeen ideas that didn't work, the prompt that kept hallucinating, the schema migration that ate a Saturday.

I'm going to do the opposite. The build is the content. The build is the proof.

There's a second reason. The gap between people who can use AI well and people who can't is about to be the widest skill gap of our lifetimes — and most of what's published about it is either hype ("AI will replace you!") or fear ("AI will replace you!"). Neither helps anyone actually build anything.

I'd rather just show you. Here's what I'm building. Here's how it works. Here's where it breaks. Here's what I'm changing next.

The eight agents and their target dates.

• WRITER-02 drafts essays from a topic. Target: May 22.
• CRITIC-04 scores every draft against a 7-axis rubric I wrote. Anything below the bar goes back. Target: May 22.
• SCOUT-01 mines what my audience is actually asking about — Reddit threads, search trends, the questions in my DMs. Target: May 22.
• CURATOR-06 keeps this dashboard live from real data instead of mocks. Target: May 29.
• PRODUCER-03 turns scripts into video — voiceover, b-roll, captions. Target: June 5.
• MATCH-07 triages inbound — sponsor inquiries, collabs, brand fit scoring. Target: June 5.
• PUBLISHER-05 schedules and ships — but only what I've approved. Target: June 12.
• ANALYTICS-08 writes the weekly transparency report — what shipped, what worked, what I changed. Target: June 12.

That's the plan. Eight agents, six weeks, one operator. The dashboard ticks forward as each one comes online.

Here's what I'd never hand to an agent.

The judgment. Always the judgment.

An agent can draft an essay. It cannot decide whether the essay is true — whether it reflects something I actually believe, learned, or changed my mind about. It can suggest a hook. It cannot know which hook would embarrass me in a year.

The model isn't "AI replaces me." The model is "AI gives me eight more hours, and I spend all eight on the 5% only I can do."

Anyone can generate content now. What can't be automated is taste, judgment, and a point of view earned the slow way. Those stay human. Mine stay mine.

That's the rule I'm building the whole system around. Every agent drafts. I approve. Nothing ships without me.

Subscribe to follow the build.

Every week, an essay breaking down one more piece of the system — the prompt I locked, the schema I refactored, the assumption I had to throw out. Plus a transparency report at the end of each month: what shipped, what didn't, what I changed.

No course, no upsell, no "DM me to learn the secret." Just the actual build, in public, while I'm still figuring it out.

The dashboard is wiring up. The first agent goes live May 22.

I'll see you in next week's drop.

— Mir