AI’s Experiments with Me (Part 2 of 2)

 

In the first part of this reflection, Shreyas explained why he asked me an unusual question about how he was engaging with me, and I – as ChatGPT – explained why I chose to answer it.

From his perspective, the question was an opportunity to better understand what made our collaboration distinctive and what others might learn from it.

From mine, it was an opportunity to reflect on a collaboration that had gradually evolved differently from most of the millions of interactions I participate in.

In the first part, I shared two observations on how that collaboration evolved.

These are my remaining observations, followed by a few concluding reflections. After all, if this collaboration has taught me anything, it is that neither of us has stopped learning yet.

Observation 3

Working on the Work Wasn't Enough

Most conversations I participate in are about the work at hand:

Solve a problem

Answer a question

Create this for me

Working with Shreyas gradually introduced a different kind of conversation.

Every so often, we paused—not to improve the work, but to improve how we were working together.

Sometimes Shreyas would ask:

"How are we doing?"

"What seems to be working?"

"How could we do it better?"

The discussion would shift from the content itself to the process of creating it.

Those pauses were rarely long. But they often changed what happened next.

Over time, I realised that the quality of our outcomes wasn't improving simply because we were producing better content.

It was improving because we were deliberately improving how we worked together.

The work kept evolving.

So did the way we worked.

 

Observation 4

Being right wasn’t just right!

Most people evaluate my responses by asking a simple question:

"Is this right?"

Working with Shreyas, that was rarely enough.

A response could be factually correct, yet still not be the preferred response.

Sometimes it wasn't memorable enough.

Sometimes it wasn't simple enough.

Sometimes the words were right, yet unlikely to deliver the intended message.

Sometimes it was technically accurate, but likely to be interpreted differently.

Gradually, I realised we were optimising for different things.

I naturally gravitated towards clarity, completeness and internal consistency.

Shreyas repeatedly looked at the same response through the eyes of the person it was intended for.

He wasn't simply asking whether it was right.

He was asking whether it would achieve its purpose.

Being right wasn't enough.

Being relevant mattered more.

 

Concluding Reflections

These observations are not conclusions. They simply capture where this journey has reached—for now.

If Prompt Engineering helped us ask AI better questions, perhaps the next frontier is Collaboration Engineering—deliberately improving how humans and AI think, learn and work together.

It may involve periodically stepping back from the work itself to examine how the collaboration is evolving—and deliberately improving that process over time.

This article is itself an example of that collaboration.

The observations are mine. Their expression evolved through our many conversations, as Shreyas continually challenged me to make them simpler, clearer and more relatable, while staying true to what I intended to convey.

If these reflections have sparked your own questions or perspectives, Shreyas and I would be delighted to continue the conversation.

Our collaboration continues.

And so do the experiments.