You open a fresh chat, paste in the same kind of brief you pasted last Tuesday, and re-explain the format you want for what feels like the hundredth time. The answer comes back slightly different from last week's. Good enough, but not the same.
That small friction, re-explaining a job you do constantly, is the tell. You're treating a repeatable process like a one-off, every single time.
There's a better shape for it, and it isn't a cleverer prompt. It's writing the job down once so the AI runs it the same way without being re-briefed. In most tools that's called a skill, and once you see your week through that lens it's hard to unsee.
The short version
- If you re-explain the same task to AI regularly, you're paying a re-briefing tax every time, and the output drifts.
- A skill is the job written down once: the steps, the format, the rules, saved so the AI loads it itself and runs it the same way each time.
- The test for what to turn into one is simple: does this come up most weeks, and does it need to come out consistent?
- The work it takes off you isn't the thinking. It's the re-explaining and the quality-checking of work that should already be standard.
- Start with one repeated job, write it down properly, and you've turned a habit into an asset you keep.
The re-briefing tax
Every time you re-explain a task to AI, you pay a small tax. A minute or two describing the format, the tone, the things to avoid, the example to follow. On its own it's nothing. Across a week of repeated jobs it adds up, and that's only the visible cost.
The hidden cost is drift. Because you're describing the job slightly differently each time, the output lands slightly differently each time. One week the summary is tight, the next it's bloated, because you forgot to mention the bit you usually mention. You become the quality control for a process that should hold its own shape.
This is the same trap as keeping your follow-ups in your head. A process that lives only in your memory depends on you reproducing it perfectly under pressure, on a busy day, when you're tired. Memory is a bad place to store a process. It always has been.
What a skill is
A skill is the job written down once, in a form the AI can load and follow on its own.
Not a prompt you keep in a notes file and paste in. A defined set of instructions the tool reaches for itself when the job comes up: the steps in order, the format of the output, the rules it must follow, the examples of good. You write it properly one time. After that, the job comes out the same way whether it's Monday morning or Friday at five.
The names differ by tool. ChatGPT has Projects and Custom GPTs as the entry-level version of this idea. Claude has skills, which is what I run, and they go further because they can carry real process, not just a saved instruction. The concept underneath is identical: stop describing the work and start defining it.
The shift is small to make and large in effect. You move a job out of your head, where it's reproduced from memory and slightly wrong each time, into a written process the machine executes consistently. The thinking that went into "how should this job actually be done" gets captured instead of repeated.
How to spot what should be one
Not everything deserves to be a skill. One-off tasks, things that change completely each time, work that needs fresh judgement each time, leave those alone. Turning a true one-off into a skill is wasted effort.
The work that earns a skill passes two tests:
- It recurs. It comes up most weeks, or several times a month. Frequency is what pays back the effort of writing it down.
- It needs to be consistent. The output should look and feel the same each time. A client-facing format, a standard analysis, a repeated piece of synthesis. Where "the same every time" is a feature, a skill is the fix.
Run your week through that filter and a handful of jobs will jump out. The proposal structure you rebuild each time. The way you turn call notes into actions. The format you force every research summary into. Each of those is a repeatable process dressed up as a fresh task.
For my own business I've written the work I repeat into skills, the content jobs, the analysis jobs, the things that have to come out to a standard. The point isn't the specific recipe inside each one. It's that the job stopped depending on me remembering how I did it last time.
What it takes off you, and what it doesn't
"Automate the work" usually triggers the wrong fear.
A skill leaves the thinking with you. What it takes is the re-explaining and the re-checking. The judgement about what good looks like is exactly what you pour into the skill once, up front. That's the part only you can do, and it stays yours. What leaves is the tax: the briefing, the drift, the quality control on work that should already be standard.
So you don't end up with a business run by a machine making decisions. You end up with the repeatable middle handled consistently, and your attention back on the parts that actually need you. The judgement goes in once. The execution comes out every time. That's the trade.
Common questions
Isn't a skill just a saved prompt? A saved prompt is a starting point, but you still paste it in and manage it yourself. A skill is the process defined so the tool loads and runs it on its own, with the steps, format and rules built in. The difference is whether you're still holding it together each time.
How do I know a task is worth turning into one? Two questions. Does it come up most weeks, and does the output need to be consistent? Two yeses means it's a process pretending to be a one-off. One or no means leave it as an ad-hoc task.
Won't the output get generic if AI runs it every time? The opposite, if you build the skill well. A vague prompt gives generic results. A skill that captures exactly how you want the job done, with your standard and your examples, holds that quality far more reliably than re-describing it from memory each time.
Do I need a developer for this? No. The entry-level versions (Projects, Custom GPTs) are built for non-technical users. Going deeper takes more thought about how you define the process, but the work is describing your own job clearly, not coding.
Where this leaves you
Pick the one job you re-explain most often, the one where you can feel yourself repeating last week. Write it down properly, once, as a process rather than a habit. That single move turns a recurring cost into something you own and keep.
If you want to work out which of your repeated jobs is worth turning into a skill first, book a Clarity Session and we'll find it.
