AutoShorts — From Opportunity to Automated Product

Short-form video has become one of the most effective content formats for audience growth and engagement. However, producing it consistently remains highly labour-intensive, particularly for creators who primarily publish long-form content.

Transforming a single long-form video into multiple short clips typically requires identifying engaging moments, editing them into standalone segments, adapting them for vertical formats, and preparing them for distribution across different platforms. When performed manually, this process can consume several hours per video, creating a significant barrier to maintaining a consistent publishing cadence.

The repetitive and time-consuming nature of short-form content production presented an opportunity to explore automation. If the end-to-end workflow could be streamlined without compromising content quality, creators could significantly reduce production effort while increasing their output.

A successful solution had the potential to help creators publish more consistently while shifting their time away from repetitive editing tasks and towards higher-value creative work.

If the process of converting long-form videos into short-form content could be fully automated, then creators would be able to increase publishing frequency without increasing operational effort, resulting in sustainable audience growth and a viable automation-driven business model.

I considered the initiative successful if we achieved:

• End-to-end automation without human intervention
• Consistent generation of publishable short-form videos
• Evidence that automatically generated clips could attract viewers
• A viable path towards monetisation or commercialisation

To validate the hypothesis effectively, I deliberately constrained the product vision around a single principle: zero manual intervention. The MVP needed to autonomously discover content opportunities, generate platform-ready clips, and publish them without requiring human involvement.

This focus prevented scope creep and ensured that every capability contributed directly to testing the core assumption behind the product.

To build this quickly, I had to be pragmatic about what to build from scratch and what APIs to leverage:

• Highlight detection — I evaluated several approaches and chose Google Gemini 2.5 Flash to analyse transcripts and score segments, optimising for speed and cost.
• Transcription accuracy — I integrated OpenAI Whisper to guarantee reliable text from audio at scale.
• Programmatic editing — Existing editing APIs were either too expensive or too constrained. I chose to build a dedicated internal service to handle ffmpeg operations automatically.

Trade-off: Full automation vs assisted workflows

Many creator tools rely on human review before publishing. I intentionally chose a fully automated approach, accepting the risk of occasional quality imperfections in exchange for validating whether zero-intervention content generation could function as a sustainable system.

All predefined success criteria were achieved. The system operated autonomously, generated audience engagement, and demonstrated a potential path towards a commercially viable product.

Within the first week of operation, generated videos averaged more than 1,000 views each while requiring no manual intervention. The system automated over 30 hours of monthly production effort that would otherwise have been spent identifying highlights, editing clips, and preparing uploads.

• 1,000+ average views per video
• 30+ hours of manual work eliminated monthly
• Automated revenue generation through YouTube Shorts monetisation

These results validated the core assumption that AI-selected clips could attract viewers and generate value without requiring human involvement in the production workflow.

The platform was intentionally designed with future commercialisation in mind. Independent services responsible for scraping, transcription, editing, and publishing enabled the system to scale beyond a single creator workflow.

This architecture provides a foundation for evolving the solution into a multi-tenant SaaS offering, allowing creators to automate content production through a shared platform.

I underestimated the value of lightweight planning in highly experimental environments. While rapid iteration accelerated learning, the absence of clearly defined milestones occasionally made progress difficult to assess.

In future projects, I would establish explicit validation checkpoints early on, ensuring that assumptions are tested incrementally without sacrificing execution speed. Structured experimentation does not slow innovation—it makes it more predictable.