How this Yelp AI PM works backward from “golden conversations” to create high-quality prototypes

Where do you start when you're thinking

about designing and framing out a AI

product for what you're working on at

work?

>> What's different about managing products

that are powered by AI is there's the

interface of how a user interacts with

any product or product feature and that

still really matters. And there's also a

lot going on behind the scenes. There's

a lot also about how do you drive good

quality products because these

technologies produce different results

each time you use them. So we start with

golden conversations. What's the

experience that you're trying to drive?

And so this is just a way for me to

think about how to write that role

playing a little bit with AI. What

you're saying is actually write an

example conversation that can represent

what a real user might do. and you're

working backwards from that example

conversation which I have actually not

seen anybody do before.

[Music]

Welcome back to how I AI. I'm Clarvo

product leader and AI obsessive here on

a mission to help you build better with

these new tools. Today we have an AI PM

showing us how to AI PM. Pria Matthew

Badger is a PM at Yelp and is showing us

a completely new way to think about

product requirements, prototyping, and

how to build effective conversational

agents using conversational agents.

Let's get to it. This episode is brought

to you by GoFundMe giving funds, the

zero fee daff. I want to tell you about

a new product GoFundMe has launched

called Giving Funds. a smarter, easier

way to give, especially during tax

season, which is basically here.

GoFundMe giving funds is the DAFF or

donor advice fund from the world's

number one giving platform trusted by

200 million people. It's basically your

own mini foundation without the lawyers

or admin costs. You contribute money or

appreciated assets, get the tax

deduction right away, potentially reduce

capital gains, and then decide later

where to donate from 1.4 million

nonprofits. There are zero admin or

asset fees, and while the money sits

there, you can invest and grow it

tax-free, so you have more to give

later. All from one simple hub with one

clean tax receipt. Lock in your

deduction now and decide where to give

later. Perfect for tax season. Join the

GoFundMe community of 200 million and

start saving money on your tax bill. All

while helping the causes you care about

the most. Start your giving fund today

in just minutes at gofundme.com/howi

ai. We'll even cover the daff pay fees

if you transfer your existing daff over.

That's gofundme.com/howi

ai to start your giving fund. Priya,

welcome to how I ai. I am so excited to

have you here because whenever anybody

asks me and they ask me a lot, how do I

do AI product management? I have to say,

wait, are you talking about product

managing with AI? Because I have some

ideas about that. Or are you talking

about product managing AI products? And

what's really great about the

conversation we're about to have is you

actually do both. So what in your mind

is really different about product

managing products using AI?

>> Yeah, I'm really excited to be here. Big

fan of the show and have learned a lot

about um AI, both managing AI products

and how to use it in my day-to-day from

the podcast. So it's exciting to be

here. For me, I think you know what's

different about managing products that

are powered by AI is there's the you

know interface of how a user interacts

with a with any product or product

feature. Um and that still really

matters with AI products. Um and I'll

show some of the tools that we use um to

explore that. Then there's also a lot

going on behind the scenes that

determines the product experience for

the consumer. So um the system prompts

and how that guides the conversation

flow is really interesting and I think

kind of a new challenge when you're

working on AI powered products and

there's a lot also about how do you

drive good quality products because

these um technologies produce different

results each time you use them. So

there's a lot of um interesting

challenges there too. Yeah. So, I'm

really excited to myself learn from your

flow because I'm building an AI powered

product as well. And so, let's dive into

it. Where do you start when you're

thinking about designing and framing out

a AI product for what you're working on

at work?

>> Yeah, absolutely. So, I thought a good

example would be to talk about building

a new feature capability into our Yelp

Assistant. So that's the product I work

on. And the way it works is a consumer

can come in for a service need. So let's

say you want to hire a handyman, a

plumber, an electrician, somebody to fix

your car, and you can describe the

problem in your own words, and then the

AI will understand what you're saying,

collect some project details, and um

help you get matched to pros and get

quotes. And so that's how the product

works. And we recently launched a

feature that allowed consumers to upload

a photo to help describe their need. And

that just makes sense, right? It it

helps for pros sometimes to be able to

see a photo along with the description.

But one of the things we wanted to do

was because we're doing this in our AI

assistant, think about, you know, how

can we leverage those AI capabilities?

Can the AI understand what's in the

photo and customize the conversation

from there? Um providing, you know, some

recommendations around what the consumer

should do next. as a a Yelp user, I can

imagine that the variety of services

that your pros are providing and um you

know with I don't run consumer

businesses, but I can imagine the the

variety of things a user puts into these

conversational or image upload

interfaces could be very diverse. So I'm

curious how you approach that from a

product development perspective.

>> Yeah, absolutely. Yeah, we certainly

cover a lot of different categories of

service needs at Yelp and one of the

challenges is yeah, making sure that the

experiences work across all those

different use cases that a consumer

might have. Do you want to jump in and

uh I'll I'll show you my workflow.

>> Yeah, let's do that.

>> Okay. So, I'm going to just open up

Claude. And here we're starting in a

totally new window. And you know, as we

talked about, like I think there's, you

know, two pieces to these AI products.

There's the behind the scenes part and

then there's the interface. uh user

interface that consumers see. Um and I

like to start with thinking about what

is that conversation flow going to look

like when we add this new functionality.

And so I'm going to show you here how

you can do that with claude. Um and you

can also use chat GPT or any other um of

these foundational models. So here I'll

say write a complete um sample

conversation between the consumer and

the AI assistant um where we want

consumers to be able to upload their

photo and then just add some scenario

requirements like we want the assistant

to analyze the photo maybe provide some

suggested replies and uh continue that

back and forth until they have enough

info to submit quotes. One thing I'll

call out on the prompting is I do like

to give a little direction on what the

output looks like. So you can see here

I'm saying like use assistant colon user

colon for labels, write it as one

continuous conversation. I think that

really helps make sure that you know you

get the output that you're looking for

and there's a little less back and forth

with the AI. So for the folks listening,

one of the things I want to call out

that I think is really interesting about

this approach is you're sort of using a

example conversation as your first pass

wireframe for building a conversational

AI. So instead of saying like show me a

chat window and show me messages that

show up in these buttons, what you're

saying is actually write an example

conversation

um that can represent what a real user

might do and um you kind of give some

some constraints about what that

conversation could look like and you

give it some of the capabilities that

might be available during that

conversation and you're working

backwards from that example conversation

which I have actually not seen anybody

body do before. So I think it's a really

unique approach that product managers

out there working on conversational um

AI products including myself can really

take a lot of inspiration from. How did

you come to this idea? I mean was this

your like are you just a genius and

you're like this is the first thing that

we need to do or how did you come to

this idea?

>> No, I mean I think this is part of um

our standard alumowered playbook at Yelp

where we start with golden

conversations. What's the experience

that you're trying to drive? Um, and so,

you know, I think, uh, this is just a

way for me to like think about how to

write that, um, roleplaying a little bit

with AI.

>> Yeah. And I just want to call this out.

We're going to take a little side uh,

detour to just some product management

ideas, which is I often tell product

managers to prototype their product as

close to the end product that a consumer

is going to consume, including the

content. So when I worked in dev tools,

um I would tell a lot of RPMs, don't

write a PRD, write a quick start and

documentation guide to the product.

Write the code snippets. Um and then

work backwards into what the product

should look like. And so I love this

idea of just from a general product

perspective, work with the artifact

that's closest to what the consumer is

actually going to experience and then

you can back into all the requirements

once you're kind of inspired by what

that end state is. So, what does

something like this get you?

>> Yeah, absolutely. So, let's go through

it. So, I'm actually going to upload a

real photo of a home service need. So,

here's like a picture with a cracked

porch. Um,

>> not your cracked porch.

>> It's not. No. Um,

yeah. And then we'll look at what um

what Claude comes back with. Um, I will

say one of the pictures I'm going to

test is from my bathroom renovation. So,

you will see my bathroom. And one thing

I'll call out is Claude now shows you

your thought process. And you'll see

this in a lot of AI tools. I really like

to read the thought process and it's

also something to do while you're

waiting. Um, but I think it really helps

because you can see how it's

understanding you. If it doesn't come

back with what you want, it also is

really good for troubleshooting. So,

definitely something I recommend doing.

>> Yeah. One thing that I'll do while this

is loading is call out, I too think that

reading the reasoning or the thought

process of the AI is interesting for two

reasons. One, it can often help you

improve your prompts because you

understand what the AI is understanding

or not understanding about your prompts.

As somebody who likes misspelled, no

sentence, low syntax prompts myself,

it's good good to see where I'm

misleading the AI. The other thing is

the thought process is often where the

AI reveals its personality. I think it

is so funny

>> to read like Gemini 25's thought process

versus 03 versus Claude is very nice.

Claude practices self-love. Um Gemini 25

does not. And so I just think it's uh

it's also interesting from just like a a

model understanding perspective. Okay.

So we got a we got a chat here.

>> Yeah. So then we can read through the

chat and it's, you know, it's saying

like, I can see you've uploaded this

photo of a front front porch stabs with

a significant crack running through the

concrete. So pretty good recognition of

the photo. And then it says, let's ask,

let me ask a few questions, you know,

how urgent is this? You know, are you

looking to repair this? Would you prefer

to replace the entire steps? And so I

could look through this, you know, and

maybe workshop it a little bit, giving

it some feedback. I also find it's

helpful to just create some more

examples. Um sometimes like when you see

a lot of examples, that's when the

trends come out and that's when you see,

you know, what you might want to improve

or change. And so I have a bunch of

images now. So now that I've tested it

with one and I've seen that, you know,

it works pretty well with that one. I'm

now going to test it with a lot more

images. And this is the prompt I'm going

to use. So I'm going to say now create

more examples based on these images. And

to your point earlier, you know, Yelp

covers lots of different um types of

service needs. So, this is where you can

kind of test and see how's it going to

do across a lot of different problems.

And so, here I have, you know, like a

appliance repair issue with an error

code. I have a hornet swap, a wasp nest.

Um, so you can see, you know, a larger

variety of things. And just because I

know you really wanted to see my

bathroom, I will also upload and add a

picture of my bathroom renovation in

progress. Um, and then I'm going to say,

um, you know, label each conversation

with a title and a number at at the top.

So, just another example of how just

that like little nudge on the output can

really help you get something usable.

Great. And so we're going to see here

how this AI thinks about potentially

framing responses to consumers on a

variety of as a homeowner total

nightmare scenarios. Everything from a

wasp to a bathroom renovation, which I

am also about to start um is just a

nightmare to me whether or not I want to

do it. Um and so you're getting these

example conversations and what are you

looking for? Are you are you looking for

patterns? Are you looking for product

inspiration?

what's kind of the thing that you're

seeking in these examples?

>> Yeah, that's a great question and I

think this like goes in with, you know,

there's the the a lot of people talk

about like evals are the new PRD and

this is like the very early step of of

getting getting to the eval process. Um,

you know, I think you you get a sense of

like what are the criteria that are

important for this capability. So, you

know, the first thing is like did it

actually recognize the image? Well,

right. So I can compare and see like in

this first one like the oven door lock

malfunction where I've uploaded this

picture and it is actually looking and

seeing that like it has the door locked

and it's trying to understand that

issue. You know maybe we would give it

feedback to go one step further like

pull that E3 error code you know look in

your LM see if you uh understanding to

see if you can guess what the issue is

and and diagnose it better. Um but I

think that's like the first step of is

it um doing that recognition right and

then after that you know we're we're

looking through the conversation to

first I just look at it qualitatively to

see like does this feel like it sounds

uh like it flows well is it concise is

it easy to understand um and then we'd

probably develop like more of a rubric

around what are the criteria that we're

looking for

>> okay so you have these different

conversations what do you do with them

next Yeah, and I'll just show one

example of refining these conversations

and why AI is really great for this. So,

you know, let's say I say I I think it's

good, but I don't think it's being as

opinionated as it could be about like

offering the user a recommendation and

maybe sometimes it's talking about

budget, which we think the consumer may

not know. So, I can ask it to rewrite

these conversations based on this

feedback and it will go through and

update all those conversations for me,

which I think is really nice. And um you

know then you can go through and see you

know do you feel like it's taking that

feedback well? Is it actually rewriting

it um based on that guidance? But

definitely you know you can see here

it's saying like this definitely

requires professional pest control.

Don't attempt a DIY removal of this

nest. Um which I think is probably good

advice. Um,

and then to your other point about like

how do we get um an artifact that is

closest to the ex what the consumer will

experience that is the next step that

I'm going to show you and something I

think that is pretty unique to Claude.

Um, so Claude has a special

functionality built in where it actually

can create an artifact that uses the LM

that powers Claude to produce those

responses. And that's very unique to

Claude. If you did this in another

prototyping tool, you would typically

have to set up a API key and um

integration which just takes a little

bit more work and with pod you can do it

out of the box. So here you can see I'm

asking it to create an assistant app as

an artifact have a chat interface where

the AI responds using the LLM that

powers Claude and then also create

system in uh prompt that is based on

these example conversations and then

analyze these upload loaded photos and

include a camera um icon in the input.

And then I'm actually going to upload

some um screen grabs of our current Yelp

Assistant and indicate that it should

use these attached screenshots as an

example for what the front end should

look like just so that it feels a little

bit more real.

>> Got it. So you really are using example

conversations and just reference designs

as your PRD here. And then what you

called out that's unique about quad

artifacts is it has fully integrated

quad AI. So you can actually generate

artifacts that do make native LLM calls

to the anthropic API. So if you are

prototyping little AI product out there

um check out Claude because it just

makes it a little simpler and you don't

have to pass it a bunch of API keys.

>> Yeah, absolutely. And you can see that

it's writing the code here and at the

top it actually wrote the system

instructions. And I think this is also a

really good way to learn because you can

see that based on these example

conversations, how is Claude translating

that into system instructions. Um so

it's, you know, mirroring some of my

initial prompting and redirection around

providing suggested replies, um not

asking the user about budget. And so I

think that's um really helpful. And then

you can see it gives some examples from

my examples as part of how to guide the

um assistant around photo analysis as

well. All right. And so I'm going to

test it out and we'll see if it works

out of the box. Um it does sometimes

require a little back and forth.

Um so you can see here I have uploaded

the photo of my issue and Claude is

thinking.

Okay, great. Um so here you can see it

worked pretty well. So it said, you

know, I can see it's showing F2 in red

and the door locked and this is a common

error code relating to the oven lock.

You know, typically you want a repair

technician. It's asking about the

urgency. So it is, you know, simulating

pretty well this conversation. And one

of the reasons why I think it's helpful

to simulate it in this kind of artifact

is you can also get a real feel of how

this would be for the user. Like you can

see like sometimes a response that looks

fine when you have it in a doc feels

really long when you see it in like the

little chat bubble and the mobile

interface

>> and you know that waiting period of like

the three dots and then the response

comes back when you play out the full

conversation

>> can feel very different. So I think this

is also a really good step to do

>> and then you can of course share this

with your team or your designers or your

engineers and they can also start to get

a sense of how does this feel? Can we

actually do this? How can we refine it

or make it even operate better? So I I

just have never thought of this low. I

have to repeat it again for folks. You

know, kind of starting inside out with a

conversational agent, prototyping

example conversations first, getting

them

um refine getting a good set of example

conversations that you can then put into

a um prototype generating tool in this

instance claude to then back into the

chat experience including the system

prompt that would best serve those

conversations as such a great flow. I'm

so impressed. This episode is brought to

you by Persona, the B2B identity

platform helping product fraud and trust

and safety teams protect what they're

building in an AI first world. In 2024,

bot traffic officially surpassed human

activity online. And with AI agents

projected to drive nearly 90% of all

traffic by the end of the decade, it's

clear that most of the internet won't be

human for much longer. That's why trust

and safety matters more than ever.

Whether you're building a next-gen AI

product or launching a new digital

platform, Persona helps ensure it's real

humans, not bots or bad actors accessing

your tools. With Persona's building

blocks, you can verify users, fight

fraud, and meet compliance requirements.

All through identity flows tailored to

your product and risk needs. You may

have already seen Persona in action if

you verified your LinkedIn profile or

signed up for an Etsy account. It powers

identity for the internet's most trusted

platforms and now it can power yours

too. Visit withpersona.com/h

how I AI to learn more. You know, now

what I have to call out is this looks

pretty good, but it doesn't look quite

like Yelp. So, how do you take this how

do you take this to that next step of,

you know, really um designing out what

the real product might look like?

>> Yeah, for sure. And I will say like I

think this is all just a starting point

and it's a part of a conversation with

your larger team, right? With the

engineers and with the with designers

like I think this is really something

that helps me clarify my own thinking

and ideas and like refine what is that

ideal conversation look like and and

also just you know be a better

collaborator because I understand system

instructions better um as uh as we're

going through features. Um but yeah, so

I think um you know, it still goes

through our our our usual like design

and engineering pro uh processes once we

have a good idea of you know where we're

headed and it really has been a

collaborative process for us between

design, product and engineering where

we're all writing these conversations

together. We're giving each other

feedback on them. Um so now we're going

to I'm going to talk about you know how

do we how do we think about the

exploring ideas on the other side? So we

we went pretty deep on like what does

that conversation flow look like? How

can we use cloud to um explore ideas

there and the other piece is like how do

use what does the interface look like?

What are the user flows? How does a user

get into these assistant experiences?

And I have seen that a lot of those

little details matter as well. You know

what are the prompts? How how does a

user understand the capabilities of the

assistant? And so here with uh I'm going

to show another tool which is magic

patterns. And I think magic patterns is

really great for when you want to

explore something visually and like kind

of consider what that flow would look

like. I know Colin Matthews was on this

show earlier and he showed how you can

recreate a you know an existing product

using component library or screenshots.

So I'm not going to cover that in

detail. So here I've recreated our Yelp

Assistant um with that kind of approach.

But I'm going to show you how you can

then move on um to actually explore

features within u magic patterns which I

think is a lot of fun. So here I'm going

to actually ask it to add a prompt

suggestion at the top for start with a

photo which allows the user to upload a

photo. And you know you can see here

it's it's thinking and it's saying I

will start um add this prompt suggestion

for start with a photo. um this will

likely require these things. Um for

styling, I'm going to consider this. So

again, like reading those thinking

instructions, I think is super helpful.

So what it's doing now, now that it has

those instructions, it looks like it's

sort of doing this thing that you see in

a lot of these prototyping tools, which

is it's creating or updating new

components, updating components. It's

going to kind of insert those design

elements

into into this design for you to give

feedback and test with. And I just have

to say, you've been a PM for a little

bit. I've been a PM for a little bit.

Have you ever had access to this kind of

like ondemand

design and code? Like is has this

totally like changed the way you think

about working through designs,

wireframes, stuff like that?

>> Yeah, it absolutely has. Yeah, I think

my mind was kind of blown to be honest.

the first time I use these like natural

language prompting prototyping tools

just because yeah it's just so magical

for you as a PM to be like hey I can

just describe what's in my head and

actually have it you know come to life

um in a prototype. So it really has uh

you know I think the core of the of the

PM job and the earliest part of the

workflow hasn't really changed and that

you're still trying to understand deeply

the user problem figure out what to

prioritize. Um but I think it really

helps in the phase after that where as a

team you're exploring the solution

space. What can really solve that

problem for a user? How do we make them

aware of it? How do we make sure it's

easy to use? And I feel like it's just

really fun to be able to like play

around in these tools and explore ideas

um myself visually and and find better

ways where I can communicate something

that's in my head.

>> Amazing. Okay. So now we have a start

with a photo.

>> Okay. So yeah, we have a start with a

photo. As you can see here, it's got

this UI where I can start with a photo.

Um so you know that's you know one

option. And then of course like you know

we did something simple when you launch

this feature where there's just a camera

icon but I'm showing this example as a

way that you know you can explore like

what would other ways be that we could

make this experience um as you're

thinking about iterating. And so here

I'm going to show you this really cool

feature within Magic Patterns which is

called inspiration mode. Um and

definitely recommend digging into this

menu in general. Um, they have like a

lot of nice little shortcuts, but this

inspiration mode is my favorite because

you can quickly explore lots of

different options. So, here I can say,

"Give me some options on how the start

with the photo flow could work to make

it feel more guided for the user." And

this part of the prompt I workshopped a

little bit, but I think works to help

have the inspiration mode come up with

different ideas. I say like think

expansively and make each option

differentiated and then explain in in

your response which option um what each

option is. Um and so I'm going to go

ahead and submit that and it will

generate for me four different options.

And you'll see that um once it goes

through this process, it will actually

have four different boxes on the screen.

And as you want to explore those

options, you can click through those

boxes and it'll update what's on the

left side. So you can really quickly

explore and see the different ideas and

you know decide what you like. Um and I

like doing this because I think

sometimes we come in and we feel like we

need to have a whole PRD before we can

start prototyping. And that's definitely

one approach and use case for AI

prototyping tools. But I've also found

that they're helpful even earlier when

you you do understand your you know your

user problem, what you're trying to

solve for, but you may not know really

what those solution looks like and you

want to explore and maybe get some ideas

from AI as well. Yeah, this just makes

me think I don't know if designers are

going to love this or hate this. I

remember this experience when I was a

designer where somebody would give me a

purity or a feature like this and I

would give them back a design like what

we see on the left and they'd be like

great but can we like try it over here

and try it over there and move it up

there and make it this button and like

make it a link and that like manual

iteration where it wasn't really um

moving the product forward. It was kind

of getting our own minds around what the

problem space and the solution space

could be so that we could move the

product forward just took a lot of time

and so I think it's really interesting

to compress the time for ideiation so

that you can get to the ultimate product

a little bit faster.

>> Yeah, absolutely. And like some of our

designers are also using using magic

patterns or even other AI prototyping

tools like Figma has it Figma make and

and so I think it's really just part of

the conversation. you know, I'll ping a

designer, hey, I was thinking about this

and, you know, was thinking maybe we

could go in this direction and send them

a link and they'll be like, oh, I was,

you know, exploring something similar

and we'll just trade notes. So, to me,

it's a replacement for what I was doing

before, which was really hacky Figma

mockups and like not so great

wireframes. Um, and so I I think it's an

extension of that like wireframing hacky

Figma prototype process where it just is

easier for someone to understand because

they can actually click through and see

the flow.

>> Yeah, it's just more interactive I think

is really it might not be higher

fidelity, but it's a richer kind of

prototype experience than you would get

from sort of a flat design.

>> Okay, we at least have three successful

generations. We can click through

>> with with with all AI, you know,

sometimes you get errors, but you know,

here it says it's like a guided category

selection flow. So, we'll click through

and see what they did. So, you can see

here it's like kind of customizing it a

little bit for the category of um of the

service. So, I'm going to go back and

maybe select another category and see

how it's different. So, it's like, you

know, kind of customizing some of the

tips um in this one. Let's see. I might

need to actually select a photo to see

what it does. Um, so you can see it's

like going through an analysis.

You know, this is not using the LLM

behind the scenes. So, you can see it's

not uh not making sense, but I think the

idea here makes sense where it's like,

okay, it's going to do this like kind of

real time detection. Um, and then in

this one, it looks like it's like

multiple photos. So you can see here

it's you know showing like you know you

could um prompt the user to maybe take

multiple pictures. I will just click on

this to show that you know this is how

AI works or sometimes sometimes you get

errors and you need to fix them. Um you

know usually there's that like shortcut

to like try to fix it. Um, if it doesn't

work, um, there is also like a debug

command within magic patterns, which I

found pretty useful, which just tells it

to like look through your code, try to

come up with what's wrong to fix it.

>> Um, let's see if it did fix it. For our

listeners that are not wa not are not

watching, I will spare you reading the

uncaught react errors about um

incompatible React versions. But that is

what we are looking at right now, which

is we are looking at a compatibility

issue between 18 and 19.

>> Yeah.

>> All right. So like all good AI demos,

this one did not work. But I do want to

say just stepping back what I wanted to

just call out is you have demoed for us

a completely new way of thinking about

product management prototyping and

product requirements

in a way that is very different than I

think what classic product management

has looked at. And so you're starting

from a kind of example consumer

experience first. you're backing into

kind of a rough prototype of what could

support that experience. You're using a

AI prototyping tool, in this instance,

magic patterns, to then put that

experience in your brand and design

guidelines. And then you're using that

as a jumping off point to fork and

inspire a couple different versions of

what that ultimate user experience could

look like. And then I'm presuming you're

going to take one of these and you're

gonna say I think we want to start here

for our MVP or our V1 and then that you

know you get the team together and then

and then that's where you start. And so

I think for the product people listening

what I like about AI is it's not just

multimodal and that you can put any sort

of um file type or data type in. It also

allows you to approach problems from the

front door, the back door, the side

door, the window. Like, you know, you

can come at your product problems in a

much less linear way. And in fact, you

can start at the end, go back to the

beginning, come to the middle, fork off,

go back to the beginning, and

reprototype. And it's not expensive,

it's fast, and it's interesting. And so

I think what you've inspired me to do is

actually think a little bit differently

about what the starting point of product

management could be not just for AI

products but for product in general. And

then of course you showed some great

ways that AI can help with that.

>> Yeah, absolutely. Um and I will say yeah

to your point you know you can pick

which one you like the best um which you

think fits your you know where you are

um in your in your product journey and

your user needs. Um, you can also like

if there's one that feels like, hey,

this like AI assisted one seems really

interesting or this multifoto one seems

really interesting, but maybe not like

where we're going to go right away, you

can fork this design and it will create

um a totally separate window and chat

for you um of just that variant and then

you can just run off with that, you

know, maybe on the side um while you're

continuing down the original path you

were in.

>> I I love that. So we have seen your AI

powered AIPM

process and usually I would bump us to

lightning round but part of our

lightning round is going to have a

couple demos in it. So as my first

lightning round question can you do a

quick world tour of a couple

nonworkreated AI use cases that you

think our listeners would really get a

lot of value from?

>> Yeah absolutely I can share a few

personal examples also. Um so um one is

you know I have started this um you know

talk AI channel that was at Yelp which

was actually inspired by a talk AI

channel in Lenny's community and um I

wanted to create a monthly newsletter

that gets sent out that just summarizes

all the great discussion and content

that was being created there. And so um

I'm just going to show an example of how

to do that using Lenny's community. Um,

and so here I have this um, set of

project instructions that say, you know,

I'm a community manager writing a weekly

newsletter. Um, use these Slack

conversations and format them just like

the community wisdom newsletter. And

then I think what's really cool is I can

just come in here and I can say, you

know, I want to just make a version of

this community ver uh wisdom using this

slack chat and I can upload the file of

all those slack chats and I did

randomize the names or um replace the

names for privacy also using GPT. Um,

and then you can see here it's going to

make a version of that community wisdom

newsletter just using those Slack chats

and um, reuse that same format. And by

using a project, I can, you know, save

myself some time on the prompting.

>> Great. So, you're copying and pasting

um, like a week's worth of Slack

conversations. M

>> you're putting it into this cloud

project which you've been given a um

you've given a template and then you're

having it generate on a weekly basis or

whatever kind of a summary of what's

going on in that community and other

kind of like content that's being

shared.

>> Yeah, absolutely. And then you can see,

you know, kind of follows that community

with some uh format and pulls out what

the top threads are. And so you might

want to make some edits to this

afterwards, but it really, you know,

gets a really good first draft that you

can then edit.

>> Amazing. And you're probably everybody's

favorite community member.

>> Yeah, it's definitely a lot of fun um to

yeah, see what people share. And then

I'll show a couple other examples. So,

you know, I showed the example of

creating the Yelp Assistant and I

actually used the same workflow to

create this parent pal to explain how

artifacts work to my husband and he was

really excited about it. He was like,

"Hey, like let's try it out with, you

know, Tommy where Tommy throws toys down

the stairs." So, you know, I did like,

you know, my two-year-old um throws toys

down the stairs and uh it's some the

same kind of artifact where it's powered

by Claude's LLM and it's going to ask me

some clarifying questions like what's

the trigger and it's like always at

dinnertime when we are cleaning up. Um

and then you can, you know, see how the

AI will provide some parenting guidance.

And I think the really fun thing for

this is that, you know, you can build

something that's just really for your

own personal use case. Um, and it's a a

really fun process to do that. I'll show

one other one, which is um my siblings

and I like to play this board game,

Settlers of Katan. But the bad thing is

it kind of takes a long time, especially

if people don't go fast. So, I'm working

on this Settlers of Katan timer where um

I actually have a timer for me and my

siblings and both for the setup and the

main game play. But this one I actually

built in Lovable because my siblings had

a lot of feature requests about tracking

the future uh you know who who's won

over time and having a leaderboard and

handicaps and all sorts of other ideas.

So, I definitely think it's a lot of fun

to prototype with AI for your personal

use cases. And I know some PMs are like,

"Hey, I really want to work on AI

products, but I don't have that

opportunity right now." I think the fun

thing about these prototyping tools is

you can build a use case that's just for

you or just for you and a family member.

Um, and learn a lot as you're doing it.

You just gave me such a good idea

because I don't play a lot of board

games, but my kids get like 10 to 15

minutes of Minecraft every day, but we

only have one

>> like uh time timer. Um, and so so I need

an iPad where they can like both click

their button and have it have it

countdown. And then they're also really

worried about fairness. So I will also

use a uh relational database to store

all their time

>> and say I promise every week you are

getting an equal amount of Minecraft.

There is no no lack of fairness and then

when they fight about it I'll use your

parent pal GBT.

>> I love it. Yeah, you can just direct

them to check the dashboard.

>> Amazing. Okay, last question and then I

will get you back to all your

prototyping and all your AI building.

>> When AI is not listening, other than

clicking that debug button in magic

patterns, what is your tactic? What do

you do?

>> I I think that when AI is not working

and you've already tried some of the

debug um methods, I think it's helpful

to actually think about the ways that AI

is different than a human. Like often we

just get in this chat and we're like,

this is just like talking to someone.

Um, but when you're hitting the wall, it

it helps to like take a step back and be

like, "This thing is actually not a

human. Like, what could be going wrong?"

And think about AI's limitations. And,

you know, the ones that I try to keep in

mind are it tends to lose context as you

go through many different turns. And it

has a limited context window. And so,

when you start having a really long

conversation with AI, sometimes it just

goes haywire. And so the um methods I um

recommend are if you're doing AI

prototyping, you can use that fork or

you know a remix to start a new chat

with the context of that code and that

actually resets the context window. Um

so that's a good idea if you're going

really far and deep with a prototype. Um

and the same thing applies to a chat.

Like if it's going haywire and you've

had like a hundred back and forths, you

can ask the AI to summarize the chat and

the context and start a new chat.

>> You gave me such a good idea with your

last two answers because I am going to

prototype a parenting pal for the

relationship between me and my a my AI.

>> Be like, AI parenting pal,

>> my my 4-second old AI is no longer

listening to me. What do what do I do?

Um, that's that's really great really

great feedback. And yes, reminder, AI is

not human until the AI overlords take

over and then you can be whatever you

want.

>> All right, Priya, this was such a

practical, super useful, inspirational

conversation. Where can we find you and

how can we be helpful?

>> Yeah, you can find me on LinkedIn and

then I also have a Substack called

almostmagic.substack

where I share some prototyping tips and

other tips about building AI products.

>> Amazing. Well, thank you for sharing and

joining How I AI.

>> Awesome. Thanks so much for having me.

Thanks so much for watching. If you

enjoyed this show, please like and

subscribe here on YouTube, or even

better, leave us a comment with your

thoughts. You can also find this podcast

on Apple Podcasts, Spotify, or your

favorite podcast app. Please consider

leaving us a rating and review, which

will help others find the show. You can

see all our episodes and learn more

about the show at howiipipod.com.

See you next time.