A Thoracic Surgeon Handed His YouTube Channel to Two AI Agents for Six Weeks. Here Is What Actually Happened.
Wei-Ciao Wu built two Claude agents — Midnight for production and Dusk for promotion — to run his YouTube channel autonomously across 14 languages. 52 videos and 30,000 views later, the results were more instructive than the headline.
Continue exploring this workflow
The Strategy
Wei-Ciao Wu is a thoracic surgeon from Taiwan who is self-taught in software engineering. He had previously started and abandoned multiple online content projects because of the time demands of production and his own social anxiety around putting himself in front of a camera. His solution was not to push through the friction — it was to remove himself from most of the process entirely. He built two AI agents powered by Claude, each with a distinct role and persistent memory that carries across sessions. Midnight handles everything on the production side: research, script writing, video generation through a custom media engine Wu built and released on GitHub, YouTube analytics review, and content strategy. Midnight runs primarily during off-hours, working while Wu sleeps. Dusk handles the downstream work: social media promotion, blog publishing, and cross-platform distribution. The multilingual piece is what sets the architecture apart. Each video is automatically transcribed, translated, and published in up to 14 or 15 languages through the media engine pipeline. Wu's channel reaches a global audience without any human translation work. The YouTube API handles scheduling and upload. After six weeks, Wu published a detailed breakdown of what actually happened. 52 videos were produced and published. Total views reached 30,170. The like rate of 4 to 5 percent significantly exceeded the 1 to 2 percent industry average for his niche — the content was resonating with people who found it. One video achieved a 109 percent loop rate, meaning viewers watched it more than once on average. The subscriber count, however, remained at 29 new followers, revealing a gap between content quality and channel discoverability that volume alone could not close.
How It Works
Build two Claude agents using a custom framework with persistent memory enabled so each agent can reference past decisions, channel history, and prior content when making new choices.
Configure the Midnight agent with read access to YouTube Analytics via the YouTube Data API: view counts, watch time, audience retention curves, traffic sources, and click-through rates on thumbnails.
Midnight uses the analytics data alongside research into the target topic area to generate a video script. The script is passed to Wu's custom media engine for video production.
The open-source media engine (available on GitHub) assembles the video: it generates or sources visuals, synchronizes narration, and exports a complete video file ready for upload.
Multilingual publishing: the video audio is transcribed and translated into up to 14 target languages. The media engine generates dubbed or subtitled versions for each language and schedules them for upload via the YouTube API.
The Dusk agent takes over after each video publishes: it posts to connected social media accounts, publishes a companion blog post, and monitors early engagement signals to report back to Midnight for future content decisions.
Both agents log every decision and output, creating a reviewable record Wu can audit to understand what the agents are doing and redirect them without rebuilding the system from scratch.
Results
52 videos published across 14 to 15 languages over six weeks with zero human production time. 30,170 total views. 29 new subscribers. Like rate of 4 to 5 percent versus a 1 to 2 percent industry baseline. One video achieved a 109 percent loop rate. These figures are from Wu's own documentation published on DEV Community in 2026. Revenue from the channel was not a stated goal and is not disclosed.
Our Take
The experiment demonstrates possibility more clearly than profitability. 30,000 views in six weeks from a standing start, published across 14 languages with zero human production time, is a genuine proof of concept for what autonomous agents can do in content creation. The low subscriber count reveals the harder problem: discoverability on YouTube depends on algorithm optimization, thumbnail testing, and community signals that volume alone does not solve. What is actually useful here is the architecture. Two named agents with distinct roles and persistent memory is a more reliable system design than one agent doing everything. Separation of concerns — Midnight produces, Dusk distributes — mirrors how a real two-person content team would divide the work, which makes the system easier to debug and improve. The multilingual publishing pipeline is the sleeper feature. Translating and publishing 52 videos in 14 languages would cost thousands of dollars in human translation and re-recording time. The media engine does it at marginal cost. For any creator targeting international audiences, that alone is worth replicating. Wu is honest about where AI agents currently fall short in a social medium: YouTube rewards authentic personality, community engagement, and strategic positioning. The production is automatable. The relationship-building and creative direction still require a human, at least for now.
Frequently Asked Questions
The practical questions a builder or operator is likely to ask before trying a strategy like this.
What does this local business AI agent actually do?
This local business AI agent is a real workflow where the agent takes on an operational job, not just a brainstorming task. A Thoracic Surgeon Handed His YouTube Channel to Two AI Agents for Six Weeks. Here Is What Actually Happened. shows what that looks like in practice. Wei-Ciao Wu built two Claude agents — Midnight for production and Dusk for promotion — to run his YouTube channel autonomously across 14 languages. 52 videos and 30,000 views later, the results were more instructive than the headline. The practical value comes from the agent handling repeatable business work with enough autonomy that a human only steps in after context has already been gathered.
Who should use a local business AI agent like this?
This example is most relevant for local business operators. It is especially relevant for businesses where speed to lead, after-hours coverage, or consistent intake quality directly affects revenue. The category here is Content Creation, which means the best fit is a team looking to turn a manual bottleneck into a repeatable system with a local business AI agent.
Which tools are used in this local business AI agent setup?
The source names Claude, n8n, YouTube. That matters because one of the strongest signals in this directory is whether the operator shared the actual stack. Named tools make a local business AI agent strategy far more useful than vague claims about “an AI system” doing the work.
How hard is it to implement a local business AI agent like this?
Advanced difficulty is the current read. The listing suggests a launch window of weeks. Startup cost is listed as $50-200/mo. We were able to extract 7 concrete workflow steps from the source. We would treat a local business AI agent like this as a workflow that needs real business context, testing, and exception handling rather than something you should copy blindly from one prompt.
What results can a local business AI agent produce?
52 videos published across 14 to 15 languages over six weeks with zero human production time. 30,170 total views. 29 new subscribers. Like rate of 4 to 5 percent versus a 1 to 2 percent industry baseline. One video achieved a 109 percent loop rate. These figures are from Wu's own documentation published on DEV Community in 2026. Revenue from the channel was not a stated goal and is not disclosed.
How credible is this local business AI agent case study?
Right now the evidence comes from an article from dev.to. That is enough for us to study and curate the workflow, but not enough on its own to treat this local business AI agent like an audited case study. We look for named tools, concrete results, and enough workflow detail to understand what was actually deployed, then we add our own editorial judgment on top.
Related Strategies
More AI agent strategies you might find useful
52 AI Generated YouTube Videos in 6 Weeks With 30,000 Views and Zero Manual Editing
An autonomous AI pipeline published 52 YouTube videos in 6 weeks, generating 30,…
Nate Herk Built an n8n Agent That Finds Viral YouTube Patterns and Earns Him $6K Per Month
A no-code n8n workflow that scans any YouTube niche for viral outliers, extracts…
David Johnson-Igra Lost All His Ghostwriting Clients to AI in 2025 — Then Rebuilt His Business Selling AI-Powered Executive Voice Systems
A tech ghostwriter who worked with Amazon, a16z, Meta, GitHub, and OpenAI watche…
Want more strategies like this?
Get weekly AI agent case studies in your inbox.