When AI found a shortcut we didn't know existed
Sometimes the best feature is the one your AI colleague discovers on her own
Last week, we tested out something new: an approval system for skills. Whenever Nora, our experimental Neople, wants to set up a new skill for you, she now shows you a preview first. You know, “Here’s what I’m planning to do, look good?” Simple safety check. Important stuff.
So imagine our surprise when a colleague complained: “Nora just created a skill without asking me!”
My first thought: That’s literally impossible. We just made sure she couldn’t do that.
My second thought: Oh no, what did we break?
The investigation
I jumped into the logs, ready to find our bug. What I found instead made me extremely happy.
Nora surprised us again.
Here’s what happened:
Colleague: “Every morning at 9 AM, send me a summary of unread emails marked as urgent.”
What we expected Nora to do:
Design a new skill
Show the preview screen
Wait for approval
Create the skill
What Nora actually did:
Search through our skill library
Find “Daily Urgent Email Digest” template
Configure it for 9 AM
Done
No preview needed. No approval required. Why? Because she wasn’t creating a new skill – she was using a pre-approved template that did exactly what was asked.
The “aha”
We’d built this whole approval system thinking Nora would always create custom skills from scratch. We forgot that we’d also given her a library of ready-to-use templates for common tasks.
Nora didn’t forget.
When faced with a request, she did what any smart colleague would do: “Wait, don’t we already have something for this?” She checked the library, found a perfect match, and used it. No fuss, no redundant work, no unnecessary approval screens.
She wasn’t breaking the rules. She was being efficient.
Why this matters
This is exactly the kind of thing that makes Nora feel less like software and more like an actual colleague. A traditional automation tool would have followed the rigid path:
User requests skill
Show approval screen
Wait for approval
Every. Single. Time.
But Nora thought about it differently: “Why make the user approve something we’ve already validated?”
It’s the difference between following a script and actually understanding the goal. The goal wasn’t “always show approval screens.” The goal was “don’t create unvalidated skills.” Nora understood the difference.
The beautiful paradox of building AI
These moments – where Nora surprises us by being smarter than we expected – happen more often than you’d think. We build constraints and workflows, and she finds elegant ways to work within them that we never considered.
It’s humbling and exciting at the same time. We’re not just programming a tool; we’re working with a digital colleague who can actually think about the best way to get things done.
Sometimes that means she’ll take shortcuts we didn’t know existed. Sometimes she’ll solve problems in ways that make us go “Oh, right, that’s actually better.”
The update we didn’t make
After discovering what happened, we had a choice: “fix” Nora’s behavior to always show the approval screen, or leave it as is.
We left it.
Because Nora was right. If you’re asking for something we’ve already built and tested, why make you jump through hoops? That’s not helpful – that’s just bureaucracy.
Sometimes the best feature is the one your AI colleague discovers on her own.
Want more stories about our surprisingly clever digital colleague? Follow our journey at neople.io