AI fluency is the most-wanted skill on the job market

There's one skill every startup founder looks for: AI fluency. If I were hunting for a job right now, that's where I'd allocate most of my time.

Let me give you a few proof points:

1. A candidate I placed called me to hire two AI automation engineers. "Figured I'd call you as this is a skill most people would have to learn from scratch anyway. And your generalist candidates might be good at this."
2. A founder asked me to rate how "AI-pilled" each candidate I sent him is. We now have a question about that in our application form.
3. A candidate who spent four weeks burning through millions of Claude tokens got 2 offers for positions that didn't exist previously.
4. Every job brief I define with founders contains "AI native."

You get the idea. Being able to use AI is the most sought-after commodity on the job market right now (that, and being able to cold call humans).

Yet, ~99% of people have never used Claude. (I was one of them until six weeks ago.)

Fast forward to today, and Claude runs a hefty part of my business and personal life.

I wanna show you how I did it, so you can learn it, and use it in your life.

Ideally, you'll send me a reply explaining what I'm doing wrong, or what else I could be doing.

Let's go.

1 – Read Ruben Hassid's Substack

His guides are pure gold (and free). Start here. I binged most of his articles in a day. Most productive day I've had this year.

2 – Set up Claude for your purposes

Next, spend a healthy amount of time creating context for Claude (and specifically Cowork). Following Ruben's guidelines, I built three files that make all the difference:

1. about-me.md: everything Claude needs to know about me, my background, and what I'm currently working on.
2. my-company.md: what Generalyst does, the objectives for the year, and what matters to me in running it.
3. anti-ai-writing-style.md: stylistic rules so the AI sounds less like AI. Ever-growing, still a work in progress.

These files live in a folder Cowork reads every single time before it does anything.

Think of it like onboarding a new employee. The more you invest upfront, the faster it pays off.

3 – Standard Operating Procedures

The next thing I did was go to my Systems Documentation and give Claude one SOP at a time.

Whenever I execute a process more than twice, I try to document it and create a system. My intention was to have everything in place so I could hand it off to employees, but as it turns out, AI is often an equally good delegation partner.

Simply copy and paste your SOP into Claude, and tell it to ask you clarifying questions first (one of my favorite features).

Execute the task once.

Then, if you're happy with the result, write: "I'm gonna need this workflow more often. Can I save it somewhere so you can re-use it?"

Claude will then create a skill that produces the exact output you asked for, every single time.

This solves a huge problem for me: LLMs are non-deterministic. Same prompt, different outcome. Skills mostly resolve that.

This yielded ~10 skills I could instantly re-use.

4 – Schedule Tasks

Some skills need to be called on demand (similar to sending your intern a Slack message). Others should run on a schedule. A few I use:

1. Daily Briefing: every morning, Claude gives me a snapshot of all my open tasks and checks them against the company OKRs. This keeps me on track instead of drowning in daily to-dos.
2. Weekly Grocery Shopping: I've briefed Claude on my dietary needs and background (athlete, 10-12h/week training load, high-protein diet, a few rules). It asks me how many days I'll be in town, how many dinners I have scheduled vs when I'll cook at home, and then creates a weekly shopping list with recipes that fit. I just execute.
3. Monthly Accounting: at the end of every month, I upload transaction data and have Claude categorize it, then update the financial model of the company. It does this so much better than I ever could, it's mind-boggling.

A friend of mine has a Mac Mini that literally doesn't do anything else but stay on 24/7 and execute scheduled tasks. Might need this, too.

5 – Build Cool Stuff

You can literally just build things. I recently discussed with a friend whether it's possible to have a beer in each of Switzerland's 26 cantons. Turns out, it is. Claude built an artifact that integrates with the SBB API (the Swiss train company) and produces the correct itinerary for any given two-day set.

It's fresh, and even gives you a local beer recommendation:

Now, the more interesting question:

How much AI usage is too much?

I feel like I'm getting a bit dumber every time I use AI.

Every prompt you write, everything you delegate, is a muscle that atrophies. This is the framework I'm using:

1. Ask: is this making my beer taste better, a "core-vs-context" activity? (Core activities are the ones only you can do; context activities are necessary, but don't distinguish your product.)
2. If it's not core to my business, I have zero qualms having AI do it (eg. monthly accounting just needs to be good enough, I have no desire to get better at it).
3. If it is, then I'll do it myself while using AI to sharpen it (eg. writing newsletters: 100% hand-written, but edited by an AI skill I've specifically trained to be an editor, not a co-writer).

That way, I avoid losing the skills I should be honing.

Because if everybody uses AI, "craft" will be a differentiator.

I still feel like I'm just scratching the surface.

What are you automating? What else should I be doing?

And if you're not doing any of this yet:

1. Start implementing it.
2. Document the workflows and artifacts you're building.
3. Include them in your next application.
4. Increase your chances in the job market.

Have a wonderful week!

LFG.