← All works
Internal Tool Design2 min read 6 sections

Zero-Cost Automation

Designing an internal automation tool around where the work already happens, the browser, so AI content and data tasks start in a click and report back on their own, on an infrastructure budget of zero.

Product
Internal AI automation suite
Role
Design and build, end to end
Interface
Chrome extensions
Constraint
$0 infrastructure
0106

The Problem

Automating AI workflows usually means paying for it. Tools like Zapier, or running the work on AWS, get expensive quickly, especially for a solo developer or a small team running complex AI tasks over and over.

The goal was a tool that made those repetitive tasks effortless to start and easy to follow, without a monthly bill.

0206

Design Goals

  1. Start work where it happens. No separate app to open and learn; tasks begin from the browser, where the content work already lives.
  2. Don't make people check. When a task finishes, the interface should know and say so, without refreshing or polling.
  3. Cost nothing. Every part of the system has to fit inside a free tier, which shapes what's possible as much as any user need.
0306

Key Decision: The Interface Is a Browser Extension

Instead of a standalone dashboard, the tool's interface is a set of lightweight Chrome extensions. People trigger a workflow directly from the page they're working on, and the extension sends the task off in one step.

It keeps the tool out of the way: no extra tab to manage and no new place to remember to look. The automation sits on top of the work rather than beside it.

0406

Key Decision: Results Report Back on Their Own

Generating video or text takes time, and the worst experience is wondering whether anything is happening. So every task is a row in a shared database, and the interface listens for changes to it.

The moment a task is picked up or finished, its row updates and the browser hears about it straight away. People queue work and carry on; the status comes to them.

0506

How It Fits Together

Three free pieces, each doing one job:

  1. Chrome extensions (where people work): the interface, where you start a task and see its status.
  2. Supabase (the shared queue): a free PostgreSQL database with real-time updates, connecting the browser to the work and back.
  3. Google Colab (the engine): free compute running the AI models and Python scripts, reachable through ngrok so new tasks can trigger it.

A task flows in a loop: queued from the browser, picked up by Colab, written back to the database, and announced to the browser in real time.

0606

The Impact

  • 5,000+ automated tasks processed in the first month.
  • 90% more content produced with the automation in place.
  • About $300/month saved in subscriptions, on an infrastructure bill of $0.