WEBVTT

00:00:00.000 --> 00:00:31.715
In the design process, sometimes the  conversation is the design.
And this idea gives me a new opportunity to remind you that we're doing an almost imaginative exercise here,
but that building products will always require collaboration.
Product design is about *people* using their critical thinking and feeling skills to solve problems for other people.

00:00:31.715 --> 00:01:07.349
And the process we're going through is not ideal, because it excludes real humans from it.
But that's the goal here: learning about how far we can go with just AI and understanding its limitations on the way.
Now we have an understanding of the market, the competition, the design questions we want to address with our product,
some insights about our fictional users, it's time for the fun part – *brainstorming*.
This is the part where I feel that AI is most robust right now, so let's see how we can co-create ideas with the help of AI.

00:01:07.349 --> 00:01:31.645
I will be feeding the 'how might we's and user stories to ChatGPT and ask it to come up with corresponding ideas to address them.
The prompt could be something like: "Amazing, what are some ideas for a product (mobile application)" – it's important to specify context –
"that could address these user needs and stories?" Let's see what our AI-powered brainstorming partner will come up with.

00:01:31.645 --> 00:02:04.825
As you can see, the initial response  is pretty generic.
However, using prompting-refinement techniques, I will feed it more context and detailed instructions to get it where I want to.
I'm going to take the most interesting ideas and ask it to dive deeper into them while keeping the persona in mind.
I could also ask it to shorten the list of ideas, but this time I choose to use my critical thinking to choose the juiciest ones.

00:02:04.825 --> 00:02:35.183
I'm going to narrow down the scope of this conversation by prompting GPT to expand on a couple of ideas:
the focus mode and smart task prioritization.
As you can see, it's pretty interesting that GPT is already listing features and functionality.
Okay, so these feel like pretty promising ideas – quite tangible.
So, the natural next step would be to put more *detail* onto them, as well as more *clarity*.

00:02:35.183 --> 00:03:02.519
What do these ideas actually mean? How would these features actually look?
Basically translating them into functionality.
With the promising ideas that GPT generated, we're going to move further down the UX process line
and try to translate them into functionality and features. This stage makes no exception for requiring critical thinking efforts.
AI is not yet in a place in which it can reliably tell us what to build.

00:03:02.519 --> 00:03:32.610
If you really want to build meaningful products made of relevant features and useful functionality,
you have to think it out yourself. AI will probably generate something quite generic, and it doesn't have the capacity
to innovate and build an interesting, surprising product.
I'm going to prompt GPT to translate the ideas mentioned before into a set of features.
I'm going to focus the conversation on one of the ideas we've identified. The prompt could be something like:

00:03:32.610 --> 00:04:00.381
"Can you translate the ideas for Focus mode into features?"
You could continue this exercise by trying to prioritize the list of features with the help of GPT.
The prompt could be something like, "Which of these are the most important ones,
the key features that would differentiate  this product from others in the market?"

00:04:00.381 --> 00:04:30.705
It feels that GPT sort of lost its way  here and it's repeating some of the earlier answers.
But with refinement and proper prompts,  we can probably get it back on track
and back in the context we were in.
So, I'm going to try  to correct its course by instructing it to use
the answers from the previous question  to create a priority list between them.  

00:04:30.705 --> 00:05:01.080
This didn't go as planned, so we will try to refine the prompt even more.
Finally, I'm getting the answer I was looking for. As you can see, I provided  
specific context, specific Instructions,  and I was pretty clear with the goal I  
had in mind when trying to get an output from GPT.

00:05:01.080 --> 00:05:34.440
It was finally able to prioritize between the feature it came up with.
Most times, you'll need a couple of prompts before you can get the best result from GPT.
This is a process of trial and error, there's a lot of experimentation involved, and I want to recommend
everyone to try a couple of prompts, try a longer conversation until you get to the place
where you feel the answer is most useful for you.
You could take this exercise even further, making it more visual with
plugins from GPT that help you generate diagrams, even user flows.

00:05:34.440 --> 00:05:59.480
Plugins are available here, and by using the list of features I've just generated and defined,
I can ask it to create a user flow.
We now have a – even though pretty basic – user journey map generated with the help of AI
based on AI content.  It's pretty exciting if you think about it.