I grew three channels to 550,000 followers organically, then I quit. Twice. Not because I ran out of ideas, but because I had a full-time job and three kids, and the content treadmill broke me. This time I did something different: I built a 7-agent AI content team with Claude Code that does about 90% of the work for under $200 a month, and below is the exact system.
If you have ever come home after a long day, sat down to record, and ended up watching other people's videos instead of making your own, you already know the real problem. It isn't ideas. It's the volume of invisible work behind every post: finding what's trending, writing a hook that lands, designing a thumbnail, editing out the filler, uploading. By the time you get through it, the camera never turns on.
So I stopped relying on willpower and built a team instead. I have no computer science degree. Six months ago I had never written a line of code. I built this by talking to Claude.
Watch the full build on YouTube, mistakes and all. Here's what I made, what it cost, and how to copy it.
Why I chose Claude Code over Claude Co-work
The first thing I did was ask Claude which tool to use. Both Claude Code and Claude Co-work run agentic workflows, so either can put AI agents to work for you. But for a multi-agent system, they are not equal.
Claude Code won for three reasons:
- It stores sensitive credentials in a
.envfile. Anything with a dot in front is a local secret only I can see, which matters when you are wiring up API keys for YouTube, Descript, and Gmail. - It handles deep API integrations without hitting a wall.
- It does the actual file work: reading transcripts, generating images, updating my Google Sheets.
Claude Co-work is more of an AI assistant who learned to code a few years ago. Claude Code is the senior developer. For seven agents, you need the senior developer.
If you are non-technical and nervous about this: Claude Code is easier to use than it looks, and once you start, you keep going. I do my best vibe coding in my pajamas, so no judgment here.
The 7 agents on my content team
I described my entire content workflow out loud using Wispr Flow (I talk faster than I type, and it understands me far better than Siri). Claude turned that ramble into a 7-agent plan in about 3 minutes.
| Agent | What it does | Tools |
|---|---|---|
| Channel Analyst | Studies my videos to learn my voice, then analyzes competitors | YouTube Data API, Claude API |
| Trend Scout | Finds daily outlier videos on YouTube and X, scores them | YouTube Data API, Rapid API, VidIQ-style scoring |
| Script Writer | Writes the hook, the 30-second scripted intro, and a bullet-point body in my voice | Claude API, brand voice doc |
| Thumbnail Designer | References winning thumbnails, generates versions in my style with my photo | Nano Banana Pro, Claude Vision |
| Video Editor | Auto-edits to the script: removes filler, adds intro captions and B-roll, matches my brand | Descript API, Underlord AI, Claude API |
| Daily Reporter | Compiles everything into Google Sheets and emails me a morning digest | Google Sheets API, Gmail API |
| Shorts Agent (Phase 2) | Repurposes long-form into shorts | Descript API |
The Shorts Agent is on the roadmap. Right now I am focused on long-form YouTube, because that is the highest-leverage format for my niche.
The outlier formula that picks my next video
Most trending lists surface the same giant creators over and over. I wanted the opposite: smaller channels whose videos are suddenly popping. That is where the untapped angles live.
My Trend Scout agent scores every video like this:
Outlier score = (views in the window ÷ the channel's average views for that window) × 100
A score above 200 is a strong outlier. Above 500 is viral. The agent prioritizes creators under 50,000 subscribers and logs the top 5 outliers per day with the title, a thumbnail screenshot, and a link.
I originally set the window to 48 hours, but even that felt too tight once I watched it run, so I moved it to 5 days. In AI, if you are a week behind, you are very behind. Five days is the sweet spot between recency and signal.
What my dashboard actually shows
The first version Claude Code built looked like generic AI-slop colors, so I handed it my brand palette (dark green, beige, brown, white) and it rebuilt the whole thing. Now I have a dashboard I actually want to open instead of a raw spreadsheet.
It shows me:
- Outlier counts from YouTube and X
- The top outlier videos, with clickable titles that jump straight to the source
- A content idea, chosen from the outliers and matched to my style
- Today's script: a fully scripted 30-second intro, then a bullet-point body
- Three thumbnail options: one that matches a proven reference, two original variations
- A competitor tracker
When I hit Start Recording, a cleaner view opens with just the bullets, no clutter. And because I sometimes share my outline on screen, Claude Code built a separate public-facing version with the structure but not the full script. Every morning at 6 a.m. I get a "Sandy AI daily digest" email with the outliers and a link to the live dashboard, because the agent has editor access to the Sheet and updates it daily.
How the AI editor actually edits
The editing agent was the hardest piece, because Descript's API is new and does not expose everything yet. My workaround was to have Claude Code write prompts for Underlord, Descript's built-in AI.
Here is what it does to a raw recording without me touching it:
- Removes filler words and repeated takes
- Starts the video exactly where I want it
- Adds captions to the intro only, not the whole video
- Pulls in B-roll and applies my intro layers and animation
- Adds subtle, quiet intro music so my voice stays front and center
- Cuts the gaps between sentences to under 0.5 seconds
It can even auto-publish to YouTube when the edit is done. I am not using that part yet because I still review every cut, but the capability is there.
What it costs (versus hiring a team)
This is the part that still surprises me.
| Tool | Monthly cost |
|---|---|
| Claude API | ~$40 |
| Rapid API (X / outlier data) | paid, usage-based |
| Descript (editing) | already paying for it |
| Nano Banana Pro (thumbnails) | subscription |
| YouTube Data API | free |
| Google Sheets + Gmail | free |
| Total | under $200 |
Now compare that to hiring: a social media manager runs $2,000 to $4,000 a month, a video editor $500 to $1,500, and a researcher $1,000 to $2,000. My system does roughly $7,000 to $8,000 of work for under $200. For a solopreneur who is just getting started, that is the difference between posting once a month and posting every day.
How to build your own
I am not going to pretend this is a 30-minute setup. It took me 5 to 6 hours spread over 2 to 3 days: about 30 minutes to get the agents outlined, and the rest on personalization, which is exactly the part that makes it yours.
The foundation is a CLAUDE.md file at the root of your project. That file becomes the brain: your channel, your brand voice, your tools, your API keys. I pasted my whole plan into it, ran /init, and Claude Code organized the rest.
If you want the shortcut, I put the full build into a Notion guide so you are customizing instead of starting from a blank page. The bones are done. You bring your niche, your brand colors, and your voice, because that last part is the whole point of a personal brand. If you want help turning this into a repeatable system, start with the Personal Brand Blueprint.
Ready to build your own content team? Grab the build guide and start with the Channel Analyst agent. Once Claude Code learns your voice, the rest of the system falls into place. Get the Personal Brand Blueprint and let Slee Studio help you automate the system, not just the posts.