AI Guides › Skills & Agents
The Difference Between An Agent And A Skill — A Plain Distinction, Finally
By Nigel Guy · 2 min read
The two words get used almost interchangeably in casual conversation, which
is a shame, because the distinction is genuinely useful once you see it: one
is a set of instructions, the other is something that acts on those
instructions with some autonomy. Confusing them leads to either
under-trusting a plain skill or over-trusting something that's actually
operating independently.
The rule: a skill is a written procedure; an agent is something that
carries out steps with a degree of autonomy, often using a skill as part of
how it works — one is the recipe, the other is the cook.
The mechanism
- Ask whether it's making decisions or following steps. A skill mostly
follows a documented procedure exactly. An agent often has to decide
which steps to take, in what order, based on what it finds along the
way.
- Ask what happens without you in the loop. A skill typically produces
an output for you to review. An agent may take several actions in
sequence before coming back to you at all — which is exactly why
permissions and scope matter more for agents.
- Match your review habits to which one you're using. A skill's single
output is one thing to check. An agent's chain of actions may need
checking at more than one point, not just at the end.
- Notice when a "skill" is actually operating like an agent — some
blur the line by chaining several autonomous steps together under a
skill's name. Read what it actually does, not just what it's labelled.
What to skip
Skip assuming the label tells you everything — the practical question is
"how much does this act on its own before I see the result," and that's
worth checking directly rather than inferring from the name.
Guardrails
- Anything acting with real autonomy deserves the write-access caution
described elsewhere in this library — start read-only, expand scope
deliberately.
- An agent's autonomy is exactly where unreviewed mistakes compound —
a small error early in a chain of actions can produce a much larger
wrong result by the end.
- The distinction matters less for casual, low-stakes use and more the
moment either one touches something real — money, external
communication, anything hard to reverse.
All 751 AI guides · JulieMango plans from £17/mo