GPT Becomes IDEO

I created an entire design thinking team from a single GPT prompt.

LLMs like GPT can do SO MUCH MORE than summarize notes or write a song about taking out the trash in the style of Slipknot (though that does have its appeal). Want proof of what’s possible inside the simple interface of ChatGPT? Just keep reading.

Introducing aiDEO: An autonomous design thinking squad

Nice pic midjourney, but who’s gonna water all those office plants?

I’ll let the team detail the process in their own words.

Not too shabby!

This team BLAZES on GPT4 but for the purposes of this demonstration, I gave myself the challenge of showing how it operates on 3.5.

First, it pulled inspiration from past innovations.

This round, the team decided to study the evolution of the smartphone. After a spirited discussion, it generated some takeaways.

Then it studied the profile of a famous innovator to learn what made them successful.

(Don’t blame me; this is who the team picked!)

They’re not wrong.

Next, the team “conducted” a social listening effort.

(Quotes because this is all a simulation since GPT3.5 isn’t connected to anything. But it still worked.)

This makes it pretty clear why platforms are trying to lock down their data. GOOD LUCK Y’ALL.

The team then used the “findings” to suggest a few problem spaces they might explore.

The model LOVES sustainability, healthcare and home automation. It also has the hots for VR.

After reviewing the potential problem spaces, the team chose the most promising one and shared a rationale for the decision.

GPT definitely has a bias towards social good. Nice work, engineers!

The team then moved on to understanding the users.

The model simulated user interviews to create three empathy maps and their corresponding personas. Here’s one of them:

We got you John!

The team took the learnings from their three user personas and identified common pain points and needs.

Next comes the fun part: IDEAS!!!

A simulated brainstorm surfaced three potential solutions to the problem.

After reviewing the potential solutions, the team selects what it deems the strongest one.

Now the team fleshes out its chosen idea.

It imagines what features this platform might include, using mind mapping as one of the strategies. The team even uses a codebox to display a simplified overview of the mind map!

Next step, it builds a prototype, including wireframes. Because the model can only generate text, the wireframes are also built inside a codebox.

Next step, user testing!

The team simulates user testing sessions and gets some helpful feedback on the prototype:

The team takes that feedback, updates the prototype and runs through a few more iterations (I’ll spare you the blow-by-blow).

The final deliverable:
A full design doc for a production-ready product.

The team tried to get sneaky and just do an outline of what a design doc might include, but after a gentle reminder, it got to work:

Yes, it does detail all of those things in the table of contents

As the very last step, the team presents its findings and asks me if I want them to continue refining the virtual telehealth platform or whether it should start all over again and discover a new problem to solve.

What was my job?

I typed “go” or “continue” or “next.” Sometimes I had to remind the team of its goal or challenge it to do more detailed thinking. I pushed it to display its work in the most powerful way. But mainly, I just watched.

My REAL job was building the prompt.

The entire process above took place inside a ChatGPT conversation, using 3.5, the free version.

It was all initiated by a single prompt.

The prompt contains 3608 characters.

Those 3608 characters were all I needed to summon an entire design thinking team that outputs a clear, detailed, actionable design doc for a real product that addresses a true user need.

Remember: This was all autonomous.

I didn’t tell the team what problem space to select. I didn’t give it any inputs during the conversation, other than to keep it on track. I didn’t feed it any social listening. I didn’t connect to a web-browsing plugin. Heck, I didn’t even use GPT4.

I did give it a goal. In this case, I challenged the team to “invent something new.”

And of course, I gave it a process to follow.

You can do this, too.

How? The details might differ, but the method is the same. The model needs three things:

a persona + a process + a goal

And if you get stuck? ASK THE MODEL. If you take nothing from this extremely long post, let it be that. The model can teach you how to work with it, if you’re willing to be curious enough to learn.

Good luck!