AI Guides › Skills & Agents
Your First Skill Should Be Boring
By Nigel Guy · 2 min read
The first skill most people try to build is ambitious — an agent that
manages their whole inbox, or automates a multi-step business process end to
end. It usually fails, not because the idea was bad, but because building
your first skill and solving a hard, multi-part problem at the same time is
two hard problems stacked on top of each other.
The rule: your first skill should automate something small, repeated, and
low-stakes — the goal of skill one is learning the shape of a good skill, not
solving your biggest problem.
The mechanism
- Pick a task you do at least weekly, not a task you do once and found
annoying. Frequency is what makes a skill worth the setup cost.
- Pick a task where a wrong output is cheap to notice and fix. A skill
that drafts a weekly status update is a good first project. A skill that
sends emails on your behalf without review is not.
- Write the "one rule" first, in one sentence, before anything else.
If you can't compress the task into one sentence, you don't understand it
well enough to automate it yet.
- Write it as a procedure you could hand to a new hire, not as a wish.
"Summarise my week" is a wish. "Pull the five items I flagged, sort by
deadline, write two lines each" is a procedure.
- Run it five times before you trust it once. Check every output for the
first handful of runs. Only then let it run with less supervision.
What to skip
Skip building anything with write access to something important (email,
calendar, financial data) as your first skill — save that for once you've
built the muscle on something with a smaller blast radius. And skip trying to
handle every edge case in version one; note the edge cases you hit and
handle them in version two.
Guardrails
- A skill that's wrong 10% of the time on a low-stakes task is a fine
learning project. A skill that's wrong 10% of the time on anything
irreversible is not ready to run unsupervised.
- Your first skill teaches you more about how you actually work than about
the tool — expect to rewrite it once you see it in practice.
- "It worked the first three times" is not the same as "it's reliable."
Small sample sizes are the most common reason people trust a skill before
it's earned it.
All 751 AI guides · JulieMango plans from £17/mo