In this episode, we catch up with Mariana Antaya, Product Manager at Microsoft, who first joined the show in February 2024. Since then, AI has gone mainstream, the tech job market has shifted, and her career has evolved significantly. Last time, Mariana shared practical advice for aspiring PMs, which became our most downloaded episode ever, and now we’re diving into what’s changed and what aspiring PMs need to succeed in 2025.
Host Hannah Clark reconnects with Mariana to reflect on the rapid developments in AI and the evolving landscape of the tech industry. They discuss how aspiring product managers can adapt and thrive in this new environment, with insights that will help listeners stay ahead of the curve.
Resources from this episode:
- Subscribe to The Product Manager newsletter
- Connect with Mariana on LinkedIn, Instagram, and TikTok
- Check out Microsoft
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Hannah Clark (00:01):
You know when you run into someone you haven't
seen in a super long timeand one of you is like, "so
what's new with you?" and it'sactually been so long since
you've seen them last that youdon't even know where to begin?
That's how I felt talkingto Mariana Antaya, who
was a guest of ours backin February of 2024.
When Mariana was on the showback then, her career as
a product manager had justbegun, she just launched her
TikTok account, and she wasgoing on her third year as the
(00:22):
founder of quantifAI, a cryptostrategy optimization tool.
But she came on with some superpractical tips for aspiring
PMs looking to break into thefield which, to this day, is our
most downloaded episode ever.
And right now, I can'teven believe that
was only a year ago.
Since then, AI has gonemainstream, the tech job market
looks completely different,and our respective platforms
(00:43):
have grown significantly.
So, we thought it was hightime to invite Mariana back
to the show to catch up, andmore importantly, catch today's
aspiring PMs up on what theyneed to know today in order
to succeed as a PM in 2025.
Let's jump in.
Oh, by the way, we holdconversations like this
every week, so if thissounds interesting to
you, why not subscribe?
Okay, now let's jump in.
(01:06):
Welcome back to theProduct Manager podcast.
We have a veryspecial episode today.
We are joined today by MarianaAntaya, who was actually
featured on our podcast alittle over a year ago now.
So for those who haven'tbeen with us that long,
I feel like such a mom, aproud mom telling this story.
Mariana's not my daughter, butI still feel very proud because
the way we met is because shereached out to us on LinkedIn.
(01:28):
She commented on just a postabout one of our episodes with
an offhanded, "I would loveto be on a podcast someday."
And Becca, our producer andI were chatting and we're
like, "let's have her on,let's see what she's about."
And it ended up being ourmost popular episode of all
time by quite a wide margin.
So Mariana is a productmanager at Microsoft now,
(01:49):
and I'm talking about herlike she's not right here.
So thank you for joining us.
Mariana Antaya (01:53):
Of course.
Absolutely.
I'm super excited toreunite with you and be
able to talk the latest andgreatest in AI and product,
and it's been a journey.
We've both grown somuch, so I'm super proud.
Hannah Clark (02:06):
Me too.
Oh my gosh.
Okay, so when we lastspoke, we were just
getting this show started.
You were just getting yourcareer off the ground.
You started doingTikTok pretty recently.
We had started recordingand putting out episodes
pretty recently.
Both of us have experienced avery massive shift in our lives.
So enough about us.
Tell us about you.
(02:27):
What's been goingon in your career?
What's been going on withyour social media content?
Tell me everything.
Mariana Antaya (02:32):
Yes, awesome.
My career has beendoing super well.
I'm still at Microsoft andwe've definitely shifted more
to an AI product focus andstrategy as most companies have
as well, which has been reallyfascinating to see that shift
really take place and growwithin, the company culture.
And then on the content side,I am creating a lot of product
(02:55):
content, but a lot of artificialintelligence content as well and
really highlighting use cases ofAI for businesses or use cases
that my audience would reallybe fascinated in day to day.
One of the most recentones I did was predicting
the race outcomes of aFormula One race, which is
a hobby that I do myself.
(03:17):
And it was like, why not justpost it out there and see
if people would resonate?
And it just seems like peopleare really resonating with
all the coding projects and.
The really cool andawesome use cases that
AI has brought to light.
Hannah Clark (03:31):
Yeah, absolutely.
And I am so excited to diginto this because it's like,
we talked a lot to folks whoare at the very top of their
game, CPOs and VPs of productwho are looking at implementing
AI through an organizationfrom a leadership lens.
And now we're really talkingto, you are really using
it and in the weeds as apractitioner day in and day out.
(03:52):
So let's talk a little bitabout the shift in culture
and AI's gonna, we'll have awhole bunch of stuff on AI,
but just in general, when youthink about how the culture of
product was when you startedin your career and now people
coming in who are more juniorthan you, what's different now?
Mariana Antaya (04:06):
Definitely there's a big shift in
the margin of where thetechnology has advanced.
Number one.
When I was just starting,I don't think ChatGPT
was even a thing.
Now we use it every singleday in our workflow.
So that's one change is thatthe actual technology has
advanced so much in sucha short amount of time.
(04:28):
And with that comes beingable to incorporate and
understand the use cases ofhow a big company can integrate
that into our workflows andactually provide those use
cases and solutions thatcustomers will wanna use.
So there's a really bigshift in culture from.
More of a holisticapproach to now.
Also a little bit more of anexperimental approach as well,
(04:52):
and trying to iterate fasterto better understand what use
cases are customers willingto pay for and actually buy,
or what use cases are actuallygoing to land in this new space.
It is definitely more of anexperimental and a faster
moving pace from when I started.
Hannah Clark (05:12):
One of the things that we talk a lot
about just as a culture andnecessarily just on the show,
is just layoffs and sort ofvolatility in the tech market.
And I'm wondering from yourperspective, do you see this
as being like a time wherepeople who are looking to
get into the profession havemore competition, or is it
more just that people arelooking for different skills?
Mariana Antaya (05:33):
I think people are going to start looking for
different and more niche typeof skills for those use cases.
If you're a domain expert, Ithink that's really gonna come
to light in this era because.
The specific prompts orthe specific use cases
that XY, Z target market isreally gonna resonate with.
So if you can really dialin on a niche that you have
(05:57):
domain expertise in, or youhave hands-on experience with,
from my perspective is what'sgonna make you like a very
successful PM in that market.
Hannah Clark (06:06):
That makes a lot of sense.
Let's move into the AI stuff.
I know that we're bothlike nine to talk about it.
We'll chunk it out intoa few different sections.
Let's first just focuson your day-to-day.
We're, we'll talk aboutbuilding with AI in a moment.
You mentioned that ChatGPTis part of your day-to-day
workflow, and I'm sure thatit's infiltrated a bunch
of different functions.
What are some of the differentways that you're using AI to
get your day-to-day work done?
Mariana Antaya (06:27):
The biggest use cases I use it as is as a
market researcher for me and todo competitive analysis for me.
So now I no longer, need towait for a researcher to get
that information for me or abig company that gets contracted
to do all of the market andcompetitive analysis, or even
myself, it takes me less timeto go infiltrate the customer
(06:52):
or the competition and kind ofsee what new features they have.
I can just prompt an LLM toconstruct a market research
paper for me to tell me whatthe gaps my product is missing
versus theirs, and also tounderstand how our product can
be ahead of the curve as well.
That's one of the, mostpractical use cases also
(07:13):
for building product specs.
AI is not going to takeaway our career as product
managers just because theycan write a product spec.
Even some of the LLMs, you stillneed that human interaction
and component because we'rethe ones physically also
talking to customers likeall day and every day.
So being able to have thathuman touch is important, but it
(07:36):
gives me a really good baselinefor the product specs that I'm
writing, whether it be a newfeature or multiple features.
So that's another reallyhigh touch use case that
I love using AI for.
Hannah Clark (07:49):
Yeah.
Okay.
I think we're findingnew ways every day.
We've had some folks comeon talking about using AI as
a way to be more effectiveat parsing data from user
interviews and different waysof kind of training it to
respond as a user persona.
Yeah, it's really innovative,the different ways to support
your workflow that justweren't possible before.
Mariana Antaya (08:09):
Even creating agents.
So creating agentsis super awesome.
I was able to create anagent to basically build
out a lot of the dashboardsthat I have for my features.
So now I'm, I no longer needto spend hours building out
dashboards to see all the usermetrics, which is something
(08:29):
that, as a product manager, weheavily rely also on the data
to back up our hypotheses.
Yeah, building agents for.
Things that you do reallyrepetitively has also
been a game changer.
Hannah Clark (08:41):
Yeah.
We did a great episode withTal Raviv a little while
ago on how to build an AIcopilot for product managers
that like, took us throughthe process and like just
hearing it from him was justlike, you can really do this.
It was, yeah it's so cool.
And just moving into the skillsets, because obviously, once
we're in the field, we'regonna be using it all the time.
But how do we prepare if we'recurrently, let's say we're
(09:03):
talking to aspiring PMs, I'msure that there's more than a
few listening who are wantingto know, what do I need to
kinda be familiar with, orwhat should I get comfortable
with before I start applyingfor jobs in which I'm going
to be using AI all the time?
Mariana Antaya (09:16):
I think one of the most relevant skills you can
learn, especially as an aspiringPM, is prompt engineering
and being able to experimentwith the different LLMs.
And what use casesare best for LLMs?
For example, like whenI'm writing code, I really
love to use the Claude.
I love using ClaudeSonnet 3.7 is awesome.
(09:38):
For writing code versus I reallyprefer something like ChatGPT
in order to do competitiveanalysis or road mapping,
or even helping me structuremy prompts for other LLMs.
Actually, I will prompt thatChatGPT, hey, for this LLM,
how should I prompt to get anoptimal result for X, Y, Z?
(10:01):
One of the best, I think,and most valuable skills
you can learn as an aspiringPM is prompt engineering
and exploring and evenbuilding your own product.
There's this whole phenomenonof like vibe coding now as
well, and that's like anamazing way if, even if you're
a non-technical PM, to reallyput your product skills to
(10:22):
the task because you canliterally just prompt an LLM,
create a product and try tobuild a go-to market strategy,
try to sell your product.
Try to go to customersand ask them about their
experience using their product.
And that's real lifeproduct experience that
you're getting that isn'treally taught in schools.
And that's reallyvaluable experience that
(10:45):
is gonna translate tothe real world as well.
So I think that's definitelyone of the key points
I wanna touch on there.
Hannah Clark (10:54):
Yeah, absolutely.
Yeah, I'm very excitedabout this kind of new
vibe, coding trend.
I'm taking a workshopin it later this week.
I'm very excited to dig into.
So we'll get into promptengineering shortly.
I wanna talk a little bitabout building AI products.
So the other side, themore customer facing thing.
You touched on vibe codingas a way to get started with
building products for customers.
But building AI productsspecifically is a whole
(11:16):
other thing that's changedin the AI or in the product
management landscape.
You're currently doingthat, so what is that like?
How can you prepare forbuilding AI products?
Mariana Antaya (11:27):
I think a big part of it is understanding
that everyone is on this journeytogether right now as well.
We're all experimenting.
We don't quite know yetat the moment, like what
are the big use cases thatpeople will really gravitate
towards and what customerswill actually are willing to
(11:47):
pay for and at what marginthey're willing to pay for that
feature or that product at.
So one, we're all in thistogether, but two is really
understanding more of theexperimenting with models.
I think there are somany different models at
our disposal as a PM andcollaborate really closely with
engineering so that they makesure that accuracy is really
(12:10):
important or latency as well.
We're seeing that now thatthere's more complexities within
softwares that's going to.
Increase the time thatthings appear on the
screen, for example.
And so really understandingthe risk to reward ratio
of how long is your userwilling to wait there to
get that additional benefitto them will be important.
(12:32):
And that's the role of theproduct manager to step in
and really analyze as well.
Keeping a really close touchwith your customer will
be really important there.
And understanding theirworkflow, maybe a four or
five second delay for them tosee some extra AI generated
summary doesn't work fortheir workflow, so being
able to understand thosechallenges or really important.
Hannah Clark (12:56):
Yeah, that's a good point.
Let's move on toprompt engineering.
You mentioned that thisis like a key skill.
I think this is not just akey skill for our PMs who
are looking to, or folks whoare aspiring to become PMs.
This is like a new skill setfor everybody and I don't think
anyone can be too good at it.
So what's your like promptengineering 101, like when
you're thinking about buildinga quality prompt, what's like
(13:18):
the main thing to keep in mind?
Mariana Antaya (13:20):
The more information I say and the more
context that you can writein the prompt, the better.
If you're very vague withyour prompt, you're most
likely going to get a veryvague response in return.
So being able to really eithershow your reasoning step by step
is a great way to prompt better.
So just say, Hey, analyzeX, Y, Z, and step one, step
(13:44):
two, step three, or evenformatting your prompts
can help in that respect.
There's also.
Newer reasoning promptsthat have recently come out.
So understanding what typeof model you want to use
for your prompt will also bepretty important, but being
able to even ask the promptfor additional details.
(14:08):
Don't be afraid to tell theLLM, like it did a bad job.
Correct the LLM inthat case in order.
To get the prompt that you want.
Or like I mentioned earlier,ask a different LLM, Hey,
how am I able to get the mosteffective output of this LLM?
And there are people who havewritten papers on that as well.
(14:29):
So definitely being able toexplain and structure your
prompt as well as give it asmuch context as you can possible
all of the documents, add inimages, add in the books, add
in whatever you need into thebrain so that it can understand
where you're coming from andthe context that you need.
(14:50):
So I think that's one of themost efficient and effective
ways that I've been able to getbetter at prompt engineering
and understanding also that.
You have to tell the LLM.
Tell ChatGPT, think of yourselfas a data scientist or think
of yourself as a senior productmanager at x, Y, Z company.
(15:12):
So giving it an identity isalso a great way to structure
your prompts that way it startsthinking more so in that vein.
Hannah Clark (15:21):
Moving into iteration on stuff, 'cause you
mentioned, correcting LLMs.
Something that I've beengetting in the habit of doing
that's really been servingme well is closing a feedback
loop with them, where if itgives me an output and I make
modifications to the outputbefore I use it, then I'll
give it back here's what I'vedone with your last output.
Please remember this sothat you can give me like a
more efficient output or amore accurate output later.
(15:43):
And that kind of tends to nudgeit in the right direction, which
I've found to be really usefuland an easy way to do that.
But what kinds of tips haveyou picked up or little
like tricks up your sleevefor getting better and
better outcomes every time.
Mariana Antaya (15:57):
I think I'm still on that journey and path
as well, besides giving itmore context, like the tips.
I shared, I'm still learning.
I'm not an expert inprompt engineering myself.
I definitely just try to iterateon the response, or sometimes
I will try to add in the sameprompt into a different LLM
(16:18):
and see what the response is.
Playgrounds areespecially useful.
For this because now there'sa tool I use, or even in
GitHub and NES has this aswell, where you can compare
the output of two differentmodels side by side.
So that's a really great wayto better analyze which LLM
(16:38):
is giving you a more accurateresponse or a response that's
more tailored to what you want.
Hannah Clark (16:44):
Oh, that's a really great tip.
Yeah, I suppose that's how youcan discern like what you know,
whether Claude is the better fitfor a specific task or ChatGPT.
I was actually wondering aboutthat, was there a reason,
are there nuances to why youdecided to use ChatGPT for
more like analysis versusClaude for writing code?
And did you notice likeone thing particularly
in those specific usecases versus the others?
Mariana Antaya (17:06):
Yeah, for sure.
So for example, in the codingexample, if I'm trying to code
with ChatGPT, a lot of timesif I get an error response or
if I get an error after I runthe code and then I enter the
error into ChatGPT, sometimesit like won't fix the error.
It'll take multiple triesfor the LLM to fix the air
(17:29):
versus Claude, I have amuch higher success rate.
I just find that itwrites a lot cleaner code
and more concise code.
So that's why I prefer Claude.
But see, these are littlenuances that unless you
really try and experimentwith LLMs ones you don't know.
Hannah Clark (17:46):
It's interesting 'cause I'm
finding the same thing.
I'm almost finding thatClaude is a more of a
creative thinker if that's.
I feel a little weird usingcreativity in the same
vein as LLMs 'cause it'sso like un comfy territory.
But I do find that what I'veused it for, it's more effective
for using or for developingmarketing collateral versus
ChatGPT I've, yeah, I've agree.
(18:08):
I've found it to be a littlebit more efficient or just give
better outputs when it comes toresearch or using it for more
kind of an analytical work.
Why do you prefer ChatGPT foron that same vein, like why have
you found that ChatGPT is betterfor more analytical tasks?
Mariana Antaya (18:25):
I find that the research it can
do is very in depth aswell, a very in depth two.
It is more accurate and theway the research is outputted
is very customizable.
I can tell it, Hey, I wannachart or give me a paper.
It will more so likereason through what
(18:48):
I'm prompting it to do.
So it's also a bit of personalpreference for your use case
and for what you are going touse it for and your role too.
There's definitely a personalpreference in there makes it.
Hannah Clark (19:04):
Yeah, I definitely agree.
We were actually talkingabout this internally and a
lot of us were just fans ofhow Claude uses sepia tones.
Like just something simplelike that, just like this UX
little tiny shift that's it'sjust easier on the ice, so
it does make a difference.
I won't keep you too long,so thank you so much for
taking a break from yourbusy life and chatting
with us and catching up.
But for those who can't getenough and wanna chat with
(19:26):
you some more or wanna followwhat you're doing, where
can people find you online?
Mariana Antaya (19:30):
Absolutely.
You can find me onLinkedIn, Mariana Antaya,
or always on TikTok andInstagram @mar_antaya.
I post a bunch of codingproduct, AI videos on there, so.
Hannah Clark (19:45):
Cool.
Alright.
Thanks for being on here.
Mariana Antaya (19:47):
Always happy to chat and collaborate with you.
If you wanna coffee chat, opento talking, just feel free to
reach out to me on LinkedIn.
Hannah Clark (19:55):
Thanks so much.
Thank you.
Thanks for listening in.
For more great insights,how-to guides and tool reviews,
subscribe to our newsletter attheproductmanager.com/subscribe.
You can hear more conversationslike this by subscribing to
the Product Manager whereveryou get your podcasts.
You know when you run into someone you haven't
seen in a super long timeand one of you is like, "so
what's new with you?" and it'sactually been so long since
you've seen them last that youdon't even know where to begin?
That's how I felt talkingto Mariana Antaya, who
was a guest of ours backin February of 2024.
When Mariana was on the showback then, her career as
a product manager had justbegun, she just launched her
TikTok account, and she wasgoing on her third year as the
(00:22):
founder of quantifAI, a cryptostrategy optimization tool.
But she came on with some superpractical tips for aspiring
PMs looking to break into thefield which, to this day, is our
most downloaded episode ever.
And right now, I can'teven believe that
was only a year ago.
Since then, AI has gonemainstream, the tech job market
looks completely different,and our respective platforms
(00:43):
have grown significantly.
So, we thought it was hightime to invite Mariana back
to the show to catch up, andmore importantly, catch today's
aspiring PMs up on what theyneed to know today in order
to succeed as a PM in 2025.
Let's jump in.
Oh, by the way, we holdconversations like this
every week, so if thissounds interesting to
you, why not subscribe?
Okay, now let's jump in.
(01:06):
Welcome back to theProduct Manager podcast.
We have a veryspecial episode today.
We are joined today by MarianaAntaya, who was actually
featured on our podcast alittle over a year ago now.
So for those who haven'tbeen with us that long,
I feel like such a mom, aproud mom telling this story.
Mariana's not my daughter, butI still feel very proud because
the way we met is because shereached out to us on LinkedIn.
(01:28):
She commented on just a postabout one of our episodes with
an offhanded, "I would loveto be on a podcast someday."
And Becca, our producer andI were chatting and we're
like, "let's have her on,let's see what she's about."
And it ended up being ourmost popular episode of all
time by quite a wide margin.
So Mariana is a productmanager at Microsoft now,
(01:49):
and I'm talking about herlike she's not right here.
So thank you for joining us.
Mariana Antaya (01:53):
Of course.
Absolutely.
I'm super excited toreunite with you and be
able to talk the latest andgreatest in AI and product,
and it's been a journey.
We've both grown somuch, so I'm super proud.
Hannah Clark (02:06):
Me too.
Oh my gosh.
Okay, so when we lastspoke, we were just
getting this show started.
You were just getting yourcareer off the ground.
You started doingTikTok pretty recently.
We had started recordingand putting out episodes
pretty recently.
Both of us have experienced avery massive shift in our lives.
So enough about us.
Tell us about you.
(02:27):
What's been goingon in your career?
What's been going on withyour social media content?
Tell me everything.
Mariana Antaya (02:32):
Yes, awesome.
My career has beendoing super well.
I'm still at Microsoft andwe've definitely shifted more
to an AI product focus andstrategy as most companies have
as well, which has been reallyfascinating to see that shift
really take place and growwithin, the company culture.
And then on the content side,I am creating a lot of product
(02:55):
content, but a lot of artificialintelligence content as well and
really highlighting use cases ofAI for businesses or use cases
that my audience would reallybe fascinated in day to day.
One of the most recentones I did was predicting
the race outcomes of aFormula One race, which is
a hobby that I do myself.
(03:17):
And it was like, why not justpost it out there and see
if people would resonate?
And it just seems like peopleare really resonating with
all the coding projects and.
The really cool andawesome use cases that
AI has brought to light.
Hannah Clark (03:31):
Yeah, absolutely.
And I am so excited to diginto this because it's like,
we talked a lot to folks whoare at the very top of their
game, CPOs and VPs of productwho are looking at implementing
AI through an organizationfrom a leadership lens.
And now we're really talkingto, you are really using
it and in the weeds as apractitioner day in and day out.
(03:52):
So let's talk a little bitabout the shift in culture
and AI's gonna, we'll have awhole bunch of stuff on AI,
but just in general, when youthink about how the culture of
product was when you startedin your career and now people
coming in who are more juniorthan you, what's different now?
Mariana Antaya (04:06):
Definitely there's a big shift in
the margin of where thetechnology has advanced.
Number one.
When I was just starting,I don't think ChatGPT
was even a thing.
Now we use it every singleday in our workflow.
So that's one change is thatthe actual technology has
advanced so much in sucha short amount of time.
(04:28):
And with that comes beingable to incorporate and
understand the use cases ofhow a big company can integrate
that into our workflows andactually provide those use
cases and solutions thatcustomers will wanna use.
So there's a really bigshift in culture from.
More of a holisticapproach to now.
Also a little bit more of anexperimental approach as well,
(04:52):
and trying to iterate fasterto better understand what use
cases are customers willingto pay for and actually buy,
or what use cases are actuallygoing to land in this new space.
It is definitely more of anexperimental and a faster
moving pace from when I started.
Hannah Clark (05:12):
One of the things that we talk a lot
about just as a culture andnecessarily just on the show,
is just layoffs and sort ofvolatility in the tech market.
And I'm wondering from yourperspective, do you see this
as being like a time wherepeople who are looking to
get into the profession havemore competition, or is it
more just that people arelooking for different skills?
Mariana Antaya (05:33):
I think people are going to start looking for
different and more niche typeof skills for those use cases.
If you're a domain expert, Ithink that's really gonna come
to light in this era because.
The specific prompts orthe specific use cases
that XY, Z target market isreally gonna resonate with.
So if you can really dialin on a niche that you have
(05:57):
domain expertise in, or youhave hands-on experience with,
from my perspective is what'sgonna make you like a very
successful PM in that market.
Hannah Clark (06:06):
That makes a lot of sense.
Let's move into the AI stuff.
I know that we're bothlike nine to talk about it.
We'll chunk it out intoa few different sections.
Let's first just focuson your day-to-day.
We're, we'll talk aboutbuilding with AI in a moment.
You mentioned that ChatGPTis part of your day-to-day
workflow, and I'm sure thatit's infiltrated a bunch
of different functions.
What are some of the differentways that you're using AI to
get your day-to-day work done?
Mariana Antaya (06:27):
The biggest use cases I use it as is as a
market researcher for me and todo competitive analysis for me.
So now I no longer, need towait for a researcher to get
that information for me or abig company that gets contracted
to do all of the market andcompetitive analysis, or even
myself, it takes me less timeto go infiltrate the customer
(06:52):
or the competition and kind ofsee what new features they have.
I can just prompt an LLM toconstruct a market research
paper for me to tell me whatthe gaps my product is missing
versus theirs, and also tounderstand how our product can
be ahead of the curve as well.
That's one of the, mostpractical use cases also
(07:13):
for building product specs.
AI is not going to takeaway our career as product
managers just because theycan write a product spec.
Even some of the LLMs, you stillneed that human interaction
and component because we'rethe ones physically also
talking to customers likeall day and every day.
So being able to have thathuman touch is important, but it
(07:36):
gives me a really good baselinefor the product specs that I'm
writing, whether it be a newfeature or multiple features.
So that's another reallyhigh touch use case that
I love using AI for.
Hannah Clark (07:49):
Yeah.
Okay.
I think we're findingnew ways every day.
We've had some folks comeon talking about using AI as
a way to be more effectiveat parsing data from user
interviews and different waysof kind of training it to
respond as a user persona.
Yeah, it's really innovative,the different ways to support
your workflow that justweren't possible before.
Mariana Antaya (08:09):
Even creating agents.
So creating agentsis super awesome.
I was able to create anagent to basically build
out a lot of the dashboardsthat I have for my features.
So now I'm, I no longer needto spend hours building out
dashboards to see all the usermetrics, which is something
(08:29):
that, as a product manager, weheavily rely also on the data
to back up our hypotheses.
Yeah, building agents for.
Things that you do reallyrepetitively has also
been a game changer.
Hannah Clark (08:41):
Yeah.
We did a great episode withTal Raviv a little while
ago on how to build an AIcopilot for product managers
that like, took us throughthe process and like just
hearing it from him was justlike, you can really do this.
It was, yeah it's so cool.
And just moving into the skillsets, because obviously, once
we're in the field, we'regonna be using it all the time.
But how do we prepare if we'recurrently, let's say we're
(09:03):
talking to aspiring PMs, I'msure that there's more than a
few listening who are wantingto know, what do I need to
kinda be familiar with, orwhat should I get comfortable
with before I start applyingfor jobs in which I'm going
to be using AI all the time?
Mariana Antaya (09:16):
I think one of the most relevant skills you can
learn, especially as an aspiringPM, is prompt engineering
and being able to experimentwith the different LLMs.
And what use casesare best for LLMs?
For example, like whenI'm writing code, I really
love to use the Claude.
I love using ClaudeSonnet 3.7 is awesome.
(09:38):
For writing code versus I reallyprefer something like ChatGPT
in order to do competitiveanalysis or road mapping,
or even helping me structuremy prompts for other LLMs.
Actually, I will prompt thatChatGPT, hey, for this LLM,
how should I prompt to get anoptimal result for X, Y, Z?
(10:01):
One of the best, I think,and most valuable skills
you can learn as an aspiringPM is prompt engineering
and exploring and evenbuilding your own product.
There's this whole phenomenonof like vibe coding now as
well, and that's like anamazing way if, even if you're
a non-technical PM, to reallyput your product skills to
(10:22):
the task because you canliterally just prompt an LLM,
create a product and try tobuild a go-to market strategy,
try to sell your product.
Try to go to customersand ask them about their
experience using their product.
And that's real lifeproduct experience that
you're getting that isn'treally taught in schools.
And that's reallyvaluable experience that
(10:45):
is gonna translate tothe real world as well.
So I think that's definitelyone of the key points
I wanna touch on there.
Hannah Clark (10:54):
Yeah, absolutely.
Yeah, I'm very excitedabout this kind of new
vibe, coding trend.
I'm taking a workshopin it later this week.
I'm very excited to dig into.
So we'll get into promptengineering shortly.
I wanna talk a little bitabout building AI products.
So the other side, themore customer facing thing.
You touched on vibe codingas a way to get started with
building products for customers.
But building AI productsspecifically is a whole
(11:16):
other thing that's changedin the AI or in the product
management landscape.
You're currently doingthat, so what is that like?
How can you prepare forbuilding AI products?
Mariana Antaya (11:27):
I think a big part of it is understanding
that everyone is on this journeytogether right now as well.
We're all experimenting.
We don't quite know yetat the moment, like what
are the big use cases thatpeople will really gravitate
towards and what customerswill actually are willing to
(11:47):
pay for and at what marginthey're willing to pay for that
feature or that product at.
So one, we're all in thistogether, but two is really
understanding more of theexperimenting with models.
I think there are somany different models at
our disposal as a PM andcollaborate really closely with
engineering so that they makesure that accuracy is really
(12:10):
important or latency as well.
We're seeing that now thatthere's more complexities within
softwares that's going to.
Increase the time thatthings appear on the
screen, for example.
And so really understandingthe risk to reward ratio
of how long is your userwilling to wait there to
get that additional benefitto them will be important.
(12:32):
And that's the role of theproduct manager to step in
and really analyze as well.
Keeping a really close touchwith your customer will
be really important there.
And understanding theirworkflow, maybe a four or
five second delay for them tosee some extra AI generated
summary doesn't work fortheir workflow, so being
able to understand thosechallenges or really important.
Hannah Clark (12:56):
Yeah, that's a good point.
Let's move on toprompt engineering.
You mentioned that thisis like a key skill.
I think this is not just akey skill for our PMs who
are looking to, or folks whoare aspiring to become PMs.
This is like a new skill setfor everybody and I don't think
anyone can be too good at it.
So what's your like promptengineering 101, like when
you're thinking about buildinga quality prompt, what's like
(13:18):
the main thing to keep in mind?
Mariana Antaya (13:20):
The more information I say and the more
context that you can writein the prompt, the better.
If you're very vague withyour prompt, you're most
likely going to get a veryvague response in return.
So being able to really eithershow your reasoning step by step
is a great way to prompt better.
So just say, Hey, analyzeX, Y, Z, and step one, step
(13:44):
two, step three, or evenformatting your prompts
can help in that respect.
There's also.
Newer reasoning promptsthat have recently come out.
So understanding what typeof model you want to use
for your prompt will also bepretty important, but being
able to even ask the promptfor additional details.
(14:08):
Don't be afraid to tell theLLM, like it did a bad job.
Correct the LLM inthat case in order.
To get the prompt that you want.
Or like I mentioned earlier,ask a different LLM, Hey,
how am I able to get the mosteffective output of this LLM?
And there are people who havewritten papers on that as well.
(14:29):
So definitely being able toexplain and structure your
prompt as well as give it asmuch context as you can possible
all of the documents, add inimages, add in the books, add
in whatever you need into thebrain so that it can understand
where you're coming from andthe context that you need.
(14:50):
So I think that's one of themost efficient and effective
ways that I've been able to getbetter at prompt engineering
and understanding also that.
You have to tell the LLM.
Tell ChatGPT, think of yourselfas a data scientist or think
of yourself as a senior productmanager at x, Y, Z company.
(15:12):
So giving it an identity isalso a great way to structure
your prompts that way it startsthinking more so in that vein.
Hannah Clark (15:21):
Moving into iteration on stuff, 'cause you
mentioned, correcting LLMs.
Something that I've beengetting in the habit of doing
that's really been servingme well is closing a feedback
loop with them, where if itgives me an output and I make
modifications to the outputbefore I use it, then I'll
give it back here's what I'vedone with your last output.
Please remember this sothat you can give me like a
more efficient output or amore accurate output later.
(15:43):
And that kind of tends to nudgeit in the right direction, which
I've found to be really usefuland an easy way to do that.
But what kinds of tips haveyou picked up or little
like tricks up your sleevefor getting better and
better outcomes every time.
Mariana Antaya (15:57):
I think I'm still on that journey and path
as well, besides giving itmore context, like the tips.
I shared, I'm still learning.
I'm not an expert inprompt engineering myself.
I definitely just try to iterateon the response, or sometimes
I will try to add in the sameprompt into a different LLM
(16:18):
and see what the response is.
Playgrounds areespecially useful.
For this because now there'sa tool I use, or even in
GitHub and NES has this aswell, where you can compare
the output of two differentmodels side by side.
So that's a really great wayto better analyze which LLM
(16:38):
is giving you a more accurateresponse or a response that's
more tailored to what you want.
Hannah Clark (16:44):
Oh, that's a really great tip.
Yeah, I suppose that's how youcan discern like what you know,
whether Claude is the better fitfor a specific task or ChatGPT.
I was actually wondering aboutthat, was there a reason,
are there nuances to why youdecided to use ChatGPT for
more like analysis versusClaude for writing code?
And did you notice likeone thing particularly
in those specific usecases versus the others?
Mariana Antaya (17:06):
Yeah, for sure.
So for example, in the codingexample, if I'm trying to code
with ChatGPT, a lot of timesif I get an error response or
if I get an error after I runthe code and then I enter the
error into ChatGPT, sometimesit like won't fix the error.
It'll take multiple triesfor the LLM to fix the air
(17:29):
versus Claude, I have amuch higher success rate.
I just find that itwrites a lot cleaner code
and more concise code.
So that's why I prefer Claude.
But see, these are littlenuances that unless you
really try and experimentwith LLMs ones you don't know.
Hannah Clark (17:46):
It's interesting 'cause I'm
finding the same thing.
I'm almost finding thatClaude is a more of a
creative thinker if that's.
I feel a little weird usingcreativity in the same
vein as LLMs 'cause it'sso like un comfy territory.
But I do find that what I'veused it for, it's more effective
for using or for developingmarketing collateral versus
ChatGPT I've, yeah, I've agree.
(18:08):
I've found it to be a littlebit more efficient or just give
better outputs when it comes toresearch or using it for more
kind of an analytical work.
Why do you prefer ChatGPT foron that same vein, like why have
you found that ChatGPT is betterfor more analytical tasks?
Mariana Antaya (18:25):
I find that the research it can
do is very in depth aswell, a very in depth two.
It is more accurate and theway the research is outputted
is very customizable.
I can tell it, Hey, I wannachart or give me a paper.
It will more so likereason through what
(18:48):
I'm prompting it to do.
So it's also a bit of personalpreference for your use case
and for what you are going touse it for and your role too.
There's definitely a personalpreference in there makes it.
Hannah Clark (19:04):
Yeah, I definitely agree.
We were actually talkingabout this internally and a
lot of us were just fans ofhow Claude uses sepia tones.
Like just something simplelike that, just like this UX
little tiny shift that's it'sjust easier on the ice, so
it does make a difference.
I won't keep you too long,so thank you so much for
taking a break from yourbusy life and chatting
with us and catching up.
But for those who can't getenough and wanna chat with
(19:26):
you some more or wanna followwhat you're doing, where
can people find you online?
Mariana Antaya (19:30):
Absolutely.
You can find me onLinkedIn, Mariana Antaya,
or always on TikTok andInstagram @mar_antaya.
I post a bunch of codingproduct, AI videos on there, so.
Hannah Clark (19:45):
Cool.
Alright.
Thanks for being on here.
Mariana Antaya (19:47):
Always happy to chat and collaborate with you.
If you wanna coffee chat, opento talking, just feel free to
reach out to me on LinkedIn.
Hannah Clark (19:55):
Thanks so much.
Thank you.
Thanks for listening in.
For more great insights,how-to guides and tool reviews,
subscribe to our newsletter attheproductmanager.com/subscribe.
You can hear more conversationslike this by subscribing to
the Product Manager whereveryou get your podcasts.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Intentionally Disturbing
Join me on this podcast as I navigate the murky waters of human behavior, current events, and personal anecdotes through in-depth interviews with incredible people—all served with a generous helping of sarcasm and satire. After years as a forensic and clinical psychologist, I offer a unique interview style and a low tolerance for bullshit, quickly steering conversations toward depth and darkness. I honor the seriousness while also appreciating wit. I’m your guide through the twisted labyrinth of the human psyche, armed with dark humor and biting wit.