Just a quickie from me based on a few chats today at work; people see AI-powered workflows and assume there’s serious technical complexity underneath. They watch a video, see a GitHub repo, and think “wtf, that’s not for me”. It can be intimidating, and I’m not suggesting my mum could easily whip up an OpenClaw instance to help her in the kitchen, but having gradually (and 100% without setting out to do so) built my own AI-assisted knowledge management system, I want to let you in on a secret: half of it runs on the computing equivalent of “look in this folder and process the file”.

In my case it’s a bash script. If you don’t know what a bash script is, that’s fine, because that’s exactly my point. It could be a Python script, a Windows batch file, probably even a KiXtart script (how mid-90s of me!), or even plain english in a text file which you point your AI-of-choice at.

The specific tooling doesn’t matter nearly as much as people think. What matters is knowing what you want. If you can describe a workflow in plain language, you can build it. Here’s a real example: my system processes meeting transcripts by finding the oldest unprocessed file in a directory, running it through a set of instructions, saving the output, and archiving the original. Sounds technical. It’s a loop over some text files.

Below is the exact prompt I used recently to slightly change how my meeting notes were processed. At first I was just throwing AI at the whole problem end-to-end, so had a big prompt which had evolved a million times, and it got to the point where I looked at it and realised I was an idiot for overcomplicating things, and could deterministically achieve 90% of it with a script, and just throw AI at it for the last bit of insight extraction, summarisation, etc.

“I have a folder of meeting notes as text/markdown files at “path/to/my/files”. Build a script that processes them one at a time, in descending date order, summarises each one as per guidelines here (“path/to/some/instructions.md”), and moves the original to the archive folder.”

It’s such a simple task that any decent model these days will one-shot that without breaking a sweat.

A realisation I think I’m coming to is that the people who are embracing AI and making it work well for them are people who are good at communicating. Articulating a desired outcome clearly, with an iterative approach (try something, if its not quite right, understand why it failed and make a small tweak, try again, repeat).

The mistake I see people make is starting with the tools. They ask “what framework should I use?” or “what’s the best AI stack?” before they’ve asked the more important question: “what’s the workflow I actually want?” The tooling follows from the intent, not the other way around.

The 10x-Engineer

This is also why copying someone else’s exact setup usually doesn’t work. My system is built around how I think (and anyone who knows me can tell you that’s a dangerous thing to copy), the meetings I have, the decisions I need to track. Yours would be different, because your work is different. But that’s fine. The value isn’t in my specific commands or file structure. It’s in the approach of describing what you need, and letting AI give it a shot, and iterating from there.

You don’t need to be a 10x-engineer running 200 Ralph-loop-agents while writing a new javascript framework in Rust, or a Youtube-AI-hype-master (like and subscribe for links to my new online course!). You just need to be ✨clear✨ about what would actually make your work easier, and then make small tweaks over time.