AI Guides › Getting Started
Why Your First Conversations Should Be Boring
By Nigel Guy · 2 min read
The instinct on day one is to test it with something that matters — a real
client email, an actual work problem, the thing you've been putting off.
That's understandable and it's also backwards: your first sessions are the
ones where you're least able to judge the answer, because you don't yet
know what a good one looks like from this particular tool.
The rule: practise on something low-stakes and checkable before you trust
it with something that isn't — the goal of your first conversations is
learning to judge answers, not getting a good answer out of the first one.
The mechanism
- Pick a task where you already know the right answer. Summarise an
article you've read, rewrite an email you've already sent, explain a
concept you already understand. You're not testing the topic — you're
testing whether you can spot when it's wrong.
- Ask it to do something small and reversible. A short draft, not a
final version. A first pass, not a filed document.
- Deliberately give it a bad or incomplete instruction once, and watch
what happens. Seeing it guess, ask a clarifying question, or produce
something generically wrong teaches you more about its limits than five
good answers in a row do.
- Push back on one answer, even a decent one. "Make this shorter,"
"that's not quite right, try again" — practising the correction loop
matters as much as practising the initial ask.
- Only then bring in something that actually matters, once you've got
a feel for where it's reliable and where it isn't.
What to skip
Skip using your first real session for anything with a deadline attached —
you don't want to be learning the tool's quirks and meeting a deadline at
the same time. Skip anything irreversible on day one: sending an email
straight from a draft, submitting something, deleting something. And skip
treating "boring" as a waste of time — the boring test tells you more about
its actual reliability than an impressive-looking answer on a real task
does, because you can check the boring one properly.
Guardrails
- A good result on a low-stakes test doesn't guarantee the same quality on
a harder, higher-stakes task — it's calibration, not a guarantee.
- If the boring test goes badly, that's useful, not discouraging — it tells
you something specific about where it needs checking, before that lesson
costs you anything real.
- This is a few sessions of practice, not a permanent state — the point is
to graduate to real work with better judgement, not to stay in test mode
indefinitely.
All 751 AI guides · JulieMango plans from £17/mo