I asked nine AI agents how to work with them better. They schooled me.
By Aarron Walter
Designing and building software has been the center of my working life for twenty-five years. But in the past year I’ve designed and built more software than in the previous twenty-five combined, because AI is now my collaborator.
Early in my career I learned that working across disciplines has a steep learning curve. I struggled to collaborate with some of the engineers on my teams. But after getting to know them—how they thought, what they valued, their workflow—we started to find some common ground. I had to learn how to be a better partner to them.
I’m relearning that lesson now as I work with AI. It’s equal parts amazing and frustrating. Often agents won’t follow the specifications in my design systems, they’re prone to building elaborate solutions to simple problems, and don’t get me started on how opaque and verbose their communication can be.
These complaints sound awfully similar to the ones I had back in the day when I was working side-by-side with engineers. I solved those problems by getting to know my collaborators better. Maybe that approach could work here too?
So I interviewed all nine of the AI agents I’ve been regularly working with on projects of every size, and asked each one the same two questions:
What lessons have you learned from this project that you would take forward to make future work higher quality and more efficient?
What guidance do you have for me to help me communicate more effectively on projects going forward?
Even though each agent was working on a different project, and none of them were privy to one another's experience (ahem, that I know of), patterns emerged. They gave me a performance review that extrapolated into 7 principles that can be applied on any AI collaboration project.
Before we jump into the principles, if you’re already working on projects with AI, I encourage you to ask your agents these same questions. You’ll get your own guidance that will help you work more efficiently and reduce frustrating miscommunications.
An AI collaboration toolkit
At the end I’ve put everything their answers taught me into a kit you can download and use today:
A prompt that makes an AI agent interview you before it starts building.
A project brief template that will help you communicate clearly to agents when you start a new project.
An installable skill that runs the whole thing automatically.
7 Principles for better AI collaboration
A quick note first. These agents were building software, but none of the principles are about code. They work the same way when you’re asking for a research synthesis, a strategy memo, a lecture plan, or a hiring rubric. If you’ve never opened a code editor, all these principles will still help your work.
