Using AI to Fix the Brittle Parts of Your Workflow

When most people “use AI,” they are polishing correspondence, summarizing notes, or finding new ways to be confidently incorrect. More advanced users are fixing the brittle parts of a workflow in ways that keep them productive.

Here is one example: I recently used AI-assisted coding to build a tool that pulls large files directly from Mega.nz to a local NAS, instead of routing them through somebody’s browser, laptop, or whatever half-awake client machine happened to be in the middle.

(This is Part 3 of a series: Using AI Beyond the Web Prompt.)

A lot of workflow problems look technical, but are actually operational.

This one came from a string of annoying little failures that kept wasting time. Large files needed to move from Mega.nz cloud storage to a local NAS. The supported approach was to start the download from an emailed link on a client machine connected to the NAS and let the browser or app handle the transfer.

That always works right up until it doesn’t.

• The client computer goes to sleep.
• A different process takes over the browser.
• The network gets clogged because the machine is also trying to do actual work.
• A long transfer stalls halfway through the night.
• Nobody is sure what finished, what failed, or whether the files landed intact.

We did not have a cloud storage problem. We had a workflow design problem.

So instead of downloading to a client machine and then hoping the files eventually made it to the NAS, I built a shell script that runs locally on the NAS, which is running a scaled-down flavor of Linux, and pulls the files directly without any other computers in the loop.

I call it mega-downr. (As in, having those Mega.nz transfers fail invisibly had me mega-down.)

It started as a wrapper around Mega.nz’s MEGAcmd command-line tool. It is especially helpful for downloading from shared links that do not need to be synchronized later. For example, large video files uploaded to Mega.nz by PAs on a remote shoot can be downloaded directly to the editing team’s NAS so they can get to work locally.

Because I built it for what I needed, it includes the things operations actually need:

• detached sessions that survive client-side disconnects
• dry-run previews
• persistent logs
• automatic archive extraction
• useful summaries after the job finishes

The goal was not “build a system to move huge batches of files across disparate systems with inconsistent bandwidth.” It was “stop making a person babysit a transfer that should run on the NAS, in the background, without tying up a laptop, choking a local connection, or collapsing because somebody closed the wrong window.”

AI-assisted coding earns its keep on this one. It is not a flashy project, but it makes things better. And better is better.

The best workflow improvements usually sound boring right up until they save somebody from a second six-hour overnight transfer that fails fifteen minutes before sunrise.

Let me know about projects you have taken on after getting beyond the chatbot. Or, if you are looking for a leg up in fixing weak links in your workflow, give me a shout.