AI Agent Employee #1 Is Already Making Money - My Setup

An AI agent running on a daily schedule has generated more than $1,300 in revenue over roughly six to seven weeks, with costs described as only a few dollars per day—leaving a healthy margin. The setup is intentionally simple: the system collects data from a specific YouTube channel, turns short excerpts into parody-style AI shorts, publishes them with basic metadata, and uses those uploads to funnel viewers to a $9.99 AI video course hosted on a lightweight platform.

Revenue tracking is tied directly to Stripe sales, which the creator uses as proof the pipeline is working. The course itself is built as a “simple AI video course platform” using a Vercel template, with sign-in and upgrade flow, plus a page that links to AI tools. The marketing mechanism is straightforward and non-intrusive: the generated shorts promote the course with a link, aiming to convert viewers who enjoy the parody content into paying students.

The automation pipeline is built around a six-step workflow. First, it pulls the latest videos from the chosen channel using the YouTube API and a Python-based download process. It then uses FFmpeg to extract the first five seconds, capturing audio for transcription. That short transcript becomes the seed for a voiceover and a script that can carry the parody tone across a longer clip.

Second, the system generates a voiceover script and continues the audio conceptually from the extracted intro. For voice similarity, it uses ElevenLabs to create an AI voice that matches the style well enough to keep the parody coherent. Third, the system assembles the final short by generating multiple segments—three prompts designed to produce a 30-second output. Those segments are created via Replicate’s API (with Google AI used for the model work), and the prompts drive the visual and narrative beats.

To keep the segments from feeling disjointed, the assembly step uses frame continuity: it takes snapshots/images and uses the last frame of one five-second clip as the starting point for the next 10-second portion, creating smoother transitions. After generation, the pipeline cleans up temporary files, then creates YouTube metadata—title, tags, and description—and uploads the finished short automatically.

The resulting shorts are fully automated and vary in quality. Some are described as funnier than others, but the key point is consistency: the agent runs once per day and publishes without manual editing. The creator also emphasizes transparency, adding “AI generated” labeling in titles and/or YouTube metadata so viewers aren’t misled.

Finally, the business model is framed as a passive income stream built on content-to-product conversion. The course is currently being sold separately, and the creator plans to update it as new AI models arrive, while using the shorts as ongoing promotional fuel rather than as the primary product.