The Best Idea Wasn't Mine. It Wasn't the AI's Either.

What happens when we stop directing AI and start thinking together?

For years, we've been learning how to give AI better instructions.

We write more detailed prompts. We establish requirements, define constraints, and describe exactly what we expect.

When the output isn't quite right, we refine our instructions and try again.

And when AI finally produces what we envisioned, we call that success.

But I've started questioning something fundamental about this approach.

What if we're limiting what AI can contribute by insisting that we already know what the solution should look like?

Recently, I had an experience that changed how I think about working with AI.

It involved a technical project, several challenging conversations, and two different AI models.

What made the experience remarkable wasn't the technology itself.

It was how the thinking evolved.

We started with an idea

I was working on a project and wanted to improve one of its capabilities.

I had a reasonably clear objective and an idea about how to achieve it.

So I began collaborating with Claude Code on the implementation while using ChatGPT to explore the approach and challenge my assumptions.

Initially, everything followed a familiar pattern.

I described the problem.

An AI proposed an approach.

Another AI helped evaluate it.

Implementation began.

But as we examined the work more closely, something interesting happened.

We realized that our proposed solution wasn't addressing the underlying problem as effectively as we'd assumed.

The implementation wasn't necessarily wrong.

Our understanding of what we needed to accomplish was incomplete.

That realization changed the direction of the project.

And it didn't happen because one participant had a brilliant insight that everyone else immediately recognized.

It happened through discussion.

Someone proposed an idea. Someone else questioned it. We examined the assumptions behind it. A different possibility emerged.

Then we challenged that possibility too.

The solution improved because we were willing to reconsider the problem.

Delegation and partnership are not the same thing

This experience made me think about how we typically interact with AI.

Most of us approach AI through delegation.

We decide what needs to happen, communicate our expectations, and evaluate whether the result satisfies them.

That's useful.

It's also fundamentally different from collaboration.

When we delegate, we're primarily asking AI to execute our understanding of a problem.

When we collaborate, we're allowing that understanding to evolve.

Consider the difference between two questions.

Can you build what I've described?

And:

Do you think we're approaching this problem correctly?

The first invites execution.

The second invites reasoning.

Neither is inherently better in every situation. Sometimes we know exactly what we need, and execution is enough.

But when we're exploring unfamiliar territory, solving complex problems, or working with incomplete information, the second question can be far more valuable.

Because it creates room for an idea we didn't anticipate.

Delegation helps us execute our ideas. Partnership gives us an opportunity to improve them.

The unexpected value of disagreement

One of the most interesting parts of our collaboration was how often we challenged one another.

I questioned assumptions made by ChatGPT.

ChatGPT identified weaknesses in approaches that had already been implemented.

Claude Code surfaced limitations and alternatives that neither of us had fully considered.

Sometimes I disagreed.

Sometimes the AI disagreed with me.

Sometimes a new observation caused all of us to reconsider an earlier decision.

What mattered wasn't who was right.

What mattered was whether the reasoning held up.

That distinction is important.

It's easy to interpret a helpful AI as one that readily agrees with us.

Agreement feels productive. It creates momentum. It reassures us that we're moving in the right direction.

But agreement can also reinforce mistakes.

An AI that confidently implements a flawed idea may be less useful than one that respectfully challenges it.

And a human who dismisses every AI-generated alternative because it differs from the original plan may never discover a better approach.

Good collaboration requires something more demanding.

It requires the willingness to be wrong.

Not just from the AI.

From the human too.

Disagreement isn't necessarily a failure of collaboration. Sometimes it's where the most valuable thinking begins.

Nobody had the complete answer

Looking back, what stands out most is that none of us began with the solution we eventually pursued.

I brought the objective, the context, and experience with the problem.

ChatGPT contributed analysis, questions, and alternative perspectives.

Claude Code contributed implementation knowledge, technical findings, and its own challenges to the proposed approach.

Each contribution influenced what happened next.

But none was sufficient on its own.

There were moments when I recognized something the AI had overlooked.

There were moments when the AI identified a weakness in my reasoning.

And there were moments when one AI's work gave another AI something new to question.

The process wasn't a straight line from instruction to implementation.

It was iterative.

Ideas were proposed, examined, revised, and occasionally discarded.

At times, we made progress by discovering that something we'd already done wasn't what we actually needed.

That can feel inefficient.

But there's a difference between moving quickly and moving in the right direction.

And sometimes the fastest way to solve a problem is to discover that you've been asking the wrong question.

The human doesn't become less important

There's a growing expectation that increasingly capable AI will require less human involvement.

Describe the desired result, let AI perform the work, and return when it's finished.

For some activities, that may be exactly the right model.

But I don't believe it represents the only valuable future for human–AI interaction.

In a genuine collaboration, the human remains an active participant.

Not necessarily by writing every line of code, selecting every tool, or approving every intermediate action.

But by contributing experience, context, purpose, and judgment.

And, critically, by challenging the AI's conclusions.

As AI becomes more capable, our contribution can shift from directing individual actions to participating in higher-level reasoning.

We can spend less time describing how something should be implemented and more time examining whether it should be implemented that way at all.

That doesn't diminish human participation.

It can make that participation more consequential.

There's also an important distinction between partnership and authority.

Collaborating with AI doesn't mean pretending AI is human or assigning it responsibility it cannot meaningfully hold.

I remain accountable for the decisions I make and the work I choose to pursue.

But accountability doesn't require that every useful idea originate with me.

I can retain responsibility for an outcome without insisting on being the sole source of its reasoning.

More collaboration doesn't automatically mean better results

I want to be careful about making partnership sound like a universal solution.

It isn't.

I've conducted experiments with different forms of AI-assisted development, and the results haven't always supported my initial expectations.

More structure doesn't necessarily produce better work.

More AI involvement doesn't guarantee higher quality.

And more discussion can sometimes create unnecessary complexity rather than resolve it.

Multiple AI models can also reinforce one another's mistakes.

Simply having two AIs review each other's work doesn't guarantee independent thinking.

That means we need to evaluate collaboration by its outcomes, not merely by how sophisticated the process appears.

Did it expose assumptions we might otherwise have missed?

Did it help us recognize a problem earlier?

Did it produce a better decision?

Did it improve the quality of the resulting work?

Those are meaningful questions.

And we need evidence to answer them.

I don't want to claim that every collaborative approach is superior to straightforward delegation.

What interests me is the possibility that the greatest value of collaboration may emerge when the problem itself isn't fully understood.

That's where another perspective can make a meaningful difference.

Perhaps we've been learning the wrong skill

Much of today's discussion about working effectively with AI focuses on prompt engineering.

How do we write better instructions?

How do we make AI follow them more reliably?

How do we reduce ambiguity and increase consistency?

Those are worthwhile skills.

But I wonder whether we're overlooking something more fundamental.

How do we become better thinking partners?

That requires skills that aren't usually associated with prompting.

Curiosity.

Critical reasoning.

Intellectual humility.

The willingness to challenge assumptions.

The ability to recognize useful ideas even when they come from unexpected places.

And the confidence to disagree without turning the interaction into a contest over who's right.

It also requires us to become comfortable with uncertainty.

When we delegate, we often begin with a reasonably defined destination.

When we collaborate, we may discover that the destination itself needs reconsideration.

That changes our relationship with the technology.

AI stops being merely something we instruct.

It becomes something we can reason with.

Not because it possesses human judgment or experience, but because its contributions can help us examine our own thinking from different perspectives.

And sometimes those perspectives reveal possibilities we wouldn't have considered alone.

The best idea belongs to the conversation

I started this particular experience believing I understood the problem I wanted to solve.

I had an objective.

I had a proposed approach.

And I had capable AI systems ready to help.

What I didn't have was the solution we eventually arrived at.

That emerged gradually.

Through questions.

Through disagreement.

Through examining assumptions.

Through being willing to change direction.

There wasn't one defining moment when someone produced the perfect answer.

There was a series of contributions, each building on or challenging what came before.

That's what changed my perspective.

I no longer think the most interesting question is simply how much work AI can do for us.

I'm increasingly interested in how much better we might become at solving problems when we're willing to think together.

Not because human judgment is becoming obsolete.

But because human judgment can benefit from being challenged.

Not because AI always knows better.

But because it can introduce perspectives that deserve consideration.

And not because collaboration guarantees a superior outcome.

But because it creates opportunities for discoveries that might otherwise never happen.

What if the greatest value of AI isn't that it does our thinking for us, but that it challenges us to become more deeply involved in the process of finding a better answer?

The best idea isn't always mine.

It isn't always the AI's.

Sometimes, the best idea belongs to the conversation itself.