AI Guides › Skills & Agents

The Skill You Shouldn't Build

By Nigel Guy · 2 min read

Not every repeated-seeming task is worth automating. Some tasks look like they'll recur because you've done them twice, but the honest answer is that they're rare, quick, and different enough each time that a skill would spend more effort handling variation than the task itself takes to do by hand. Building it anyway feels productive. It usually isn't.

The mechanism

1. Check frequency honestly. "I might need this again" is not frequency. A pattern of actual repeats — weekly, or reliably several times a month — is. If you're guessing at future need rather than counting past occurrences, that's the first warning sign.
2. Check consistency. A skill pays off when the steps are basically the same each time. If every instance of the task has a genuinely different shape — different inputs, different judgement calls, different exceptions — you'll spend the build handling variation, and the resulting skill will need constant babysitting anyway.
3. Check the actual time saved. Estimate how long the task takes by hand, honestly, and how long writing, testing, and maintaining the skill will take. A task that takes three minutes by hand needs to recur a lot of times before a skill that took an hour to build, and needs occasional upkeep, comes out ahead.
4. Check whether "by hand" is actually the harder part, or whether it's something else — unclear requirements, a decision you're avoiding, information you don't have yet. A skill can't fix a task that's hard because of ambiguity rather than repetition; see "When To Build A Skill, And When You're Just Procrastinating On The Actual Task" for that distinction.
5. If it fails any of the above, just do the task. Not as a consolation — as the actually efficient choice. The instinct to formalise everything into tooling is itself worth resisting sometimes.

What to skip

Skip building a skill "on principle," because automating things feels like the more sophisticated approach — sophistication isn't the goal, net time and reliability are. And skip the sunk-cost trap of finishing a skill you've realised, partway through, doesn't clear this bar; stopping and just doing the task is the cheaper path from that point on.

Guardrails

• This is a judgement call, not a formula — nobody can hand you an exact number of repeats or minutes that makes a skill "worth it" for your specific situation, and this guide won't pretend otherwise.
• A task that fails this bar today can pass it later if your circumstances change — a one-off becomes a recurring need, or the tool it touches gets slower to use by hand. Revisit the decision rather than treating it as permanent.
• None of this argues against automation generally — it's specifically about the false economy of building a skill for a task that was never going to recur often enough, or consistently enough, to earn one.

All 751 AI guides · JulieMango plans from £17/mo