Running a YouTube Channel on Autopilot

What I Built

At a high level, the system scraped fresh news from the web, ranked stories based on relevance and expected usefulness, converted chosen stories into scripts and visuals, reviewed the output for policy risk, uploaded the finished videos, and published them on a schedule with minimal intervention.

At the later stages, I moved the workflow from my laptop to Cloudflare Workers, added cron jobs, and built a dashboard to monitor the pipeline. That meant the system could keep producing videos while I slept, and I only needed to step in when I wanted to improve prompts, tune quality, or shift strategy.

From an engineering perspective, this was satisfying. I had built a content factory that could run in the background. From a creator perspective, the real lessons began only after the factory existed.

Niche Matters More Than Generic Relevance

The hardest lesson was also the most obvious in hindsight: niche matters a lot. I approached the problem like an engineer. My assumption was that if AI could identify the most important or most useful news, the channel would naturally find its audience.

That did not happen. The channel was too broad. It covered useful news, but it did not stand for one specific thing in the viewer's mind. It had relevance, but not identity.

Automation should serve the niche. It should not be the niche.

The people who are truly good at this game usually go narrow. They build for one audience, one repeated need, and often one recognizable content format. That narrowness looks limiting from the outside, but it creates loyalty on the inside. If I do this again, I would choose the niche first and only then design the machine around it.

I Entered a Mature and Competitive Market

Another realization was that YouTube automation is not some untouched edge case. I entered a field that has already been explored for years by people who understand thumbnails, packaging, watch-time optimization, topic selection, and aggressive growth tactics far better than I did at the start.

I entered with a more idealistic mindset. I wanted to make useful, non-clickbait content and help people one video at a time. That instinct still matters. But idealism alone does not win inside a market where many participants optimize hard for reach and ranking.

If I want to stay idealistic and still compete, the answer is not simply to produce more. The answer is a stronger long-term strategy with better positioning and tighter economics.

Shorts Gave Dopamine. Long-Form Gave Signal.

I started with Shorts because they produced quick feedback. Many of them crossed 1,000 views on average, which felt exciting early on. That made the channel look healthier than it really was.

Later I shifted attention toward long-form. One video was only around three minutes long, but it reached roughly 2,000 views and generated close to 100 watch hours. That single result told me more about channel potential than many Shorts had.

Views are not all equal. Watch time, repeat behavior, and depth of engagement matter more.

That was the moment I realized I should have tested long-form earlier. Shorts are useful for momentum, but they do not automatically build a durable channel.

More Content Is Not Automatically Better

I tested multiple production strategies:

• batching videos from the previous 24 hours of news and scheduling them every 30 minutes
• scraping every two hours and publishing from a rolling shorter window
• batching every 6 hours instead of continuously
• combining several stories into a short summary
• turning one story into a deeper breakdown

Each strategy had different tradeoffs around freshness, volume, quality, and cost. What I eventually learned is that I was producing too much content. Engineering thinking made me treat throughput like a default win. Content does not work that way.

Flooding a channel can dilute quality, confuse the audience, and increase cost without improving retention. If I run this again, I would rather publish fewer, better-targeted videos than operate the system at maximum speed just because it can.

The First Version Was Manual, and That Was the Right Call

Before building the full system, I made the first video manually. I moved between Claude, Canva, Google AI Studio, and other tools to see whether the concept had any pull at all. My first Short got roughly 1,000 views within a few hours. That was not enough to prove a business, but it was enough to justify building a system around the idea.

After that, I automated piece by piece. The earliest system-generated videos were rough. Quality was inconsistent. But as I kept shipping, the pipeline improved: scripts got better, visuals improved, the workflow became more reliable, and eventually auto-upload and auto-publication came online.

One thing I noticed while studying the space is that many of the more experienced operators are still doing a lot of this work manually. They move between tools and tabs, generate assets in one place, edit in another, upload somewhere else, and coordinate the workflow through habit rather than fully codified automation.

That does not mean they are doing it wrong. In many cases, their manual process is already highly optimized. But my advantage is different: I can turn the whole workflow itself into software and let it run without living inside a maze of tabs.

Cost Per Video Matters More Than Most Engineers Expect

One of the most important metrics in the experiment was cost per video. For text generation, I mostly used cheaper models for light tasks and stronger models for harder reasoning. That part usually landed around $0.05 to $0.10 per video.

The expensive part was image generation. I wanted visuals that felt polished enough to support the narrative, so I used high-quality image generation and tried to reduce the burn by creating 3x3 image grids inside a single 4K asset.

At the experiment's peak, that translated into roughly $100 to $150 per month. More experienced operators usually constrain costs much harder, often by relying on subscription tooling, fewer uploads, and tighter topic selection. That changed the whole framing for me.

The real question is not whether AI can make the video. It is whether AI can make the video cheaply enough for the channel economics to make sense.

Policy Review Had to Become a First-Class System Component

One thing I am glad I implemented was a policy-aware generation and review layer. I built a workflow where AI would fetch YouTube policy guidance, incorporate those constraints into generation prompts, and then run a second review pass to flag content that might violate policies before publication.

This came later in the experiment, which means some earlier videos may have had issues I would not be comfortable with now. Once the review loop existed, I felt much better about enabling automatic publication.

• generate with policy constraints already in the prompt
• review scripts and visuals again before release
• publish automatically only after the review stage passes

Infrastructure Was Not the Bottleneck

Moving the workflow from my laptop to Cloudflare Workers, wiring up cron jobs, and building a dashboard made the whole experiment feel much more real. The system became durable, remote, and less dependent on me staying online.

But infrastructure was never the main challenge. The real bottlenecks were channel strategy, content positioning, and economics. That is a useful reminder for engineers: sometimes the hard part is not making the system autonomous. The hard part is deciding what it should produce, how often it should produce it, and for whom.

Why I Paused the Experiment

By the end, I had a clear list of ideas for version two:

• reduce unnecessary code and simplify the pipeline
• narrow the niche dramatically
• focus more on long-form video
• improve subscriber and watch-time growth
• reduce generation costs
• publish less often but with stronger editorial intent

But I paused because time is also a cost. The technical foundation was there. The strategic foundation was not yet strong enough to justify continued spend and attention. I would rather pause a working system than keep feeding it money without conviction.

Automation Is Real. A Channel Worth Returning To Is Harder.

This experiment proved that fully automated YouTube production is possible. You can scrape information, transform it into videos, review them, upload them, and run the whole thing in the background.

What is harder is building a channel that deserves to grow. If I revisit this idea, I would approach it less like a generic automation engineer and more like a media operator: one niche, one audience, one editorial identity, and one cost structure that actually holds up.

I still think there is real opportunity here. The opportunity is not in blindly automating content production. It is in combining strong engineering with sharp taste, strategic restraint, and a clear understanding of what people actually return for.