Check out the conversation on Apple, Spotify and YouTube.
A comprehensive guide to building AI-first products from Google’s Director of AI Product. Learn how Jaclyn Konzelmann masters Imagen, Veo, and Opal, uses three critical frameworks to ship AI products, and reveals exactly what Google looks for when hiring AI PMs.
Here is the transcript:
Introduction (0:00:00 – 0:01:38)
Aakash: We’re going to cover everything you need to become an AI PM as well as demo how to use all of Google’s AI products like a pro. Google went from way behind in the AI race to a leader. Polymarket puts their odds at having the best AI model by the end of the year at 72%. That’s because Imagen is already the best image model, Veo is one of the best video models, and with tools like Opal and Mixboard, they’re allowing you to chain together workflows into really powerful use cases. That’s why today I brought in director of AI product at Google, Jaclyn Konzelmann. Jaclyn, welcome to the podcast.
Jaclyn: Thank you. Excited to be here.
Aakash: So people keep saying the AI PM role is hype. Is it real?
Jaclyn: I think it’s absolutely real. I am a product manager who works exclusively with AI-based and AI-native products. So no, it’s definitely real.
The AI PM Role at Google (0:01:38 – 0:03:03)
Aakash: How much are AI PMs paid? Levels.fyi is showing these pretty high numbers for Google product managers. Are these accurate for Google AI PMs?
Jaclyn: I mean I think you can look at all the job postings that we have online right now and be able to easily compare the fact that AI product managers and product managers in general—they carry a lot of experience and are good at what they do—is a well-paid industry.
Aakash: For these different levels, like what is an L6 senior PM at Google, because in my experience, what might be a senior PM at a Series B startup versus a senior PM at Google can be dramatically different.
Jaclyn: I think that’s true. I joined Google what, 8.5 years ago as an L5 PM back then and I think I was a senior or group PM at the more mature startup that I was at previous to that. So you need to look at years of experience and realize that like calibrating different levels changes based off of if you’re at a startup versus at a more mature company. I will say that straight out of school, you tend to start as either an L3 or an L4 PM and then you can continue to sort of rise with more years of experience, more ships, more product experience under your belt from there.
How AI Changed Product Building (0:03:03 – 0:06:10)
Aakash: This role is very real. It’s all about building AI products. Can you give us a masterclass in building AI products? How has AI changed product building?
Jaclyn: So I think there’s a couple ways that AI has changed product building. One is in how you actually build products—how you can use new AI-native tools to get things done. And then the other is in the types of products that you do actually want to build and how does AI functionality change inherently the types of capabilities and features you want to be thinking about.
And in saying all of that, you know, I think it’s important to just really call out that it does really feel like there’s never been a more exciting time to build than right now. But with that, I’ve also noticed that it also can feel like there’s never been a more overwhelming time to build than right now. And that’s because the pace of AI is accelerating. More powerful models are coming out almost every day it feels. And that’s leading to better tools. Better tools to help you build, better tools to help you understand what’s possible to build as a product. And as a result, more and more products are coming out into the world.
And so that leads a lot of people to also sometimes feeling like there’s never been a more overwhelming time to build than right now. And I think it’s helpful to just acknowledge that. But more than anything, it’s just such an exciting time to be building because the possibilities of how to build and how fast you can ship have never been more realized than they are right now.
Aakash: 100%. So how do you build zero to one AI products?
Jaclyn: You know, it’s funny. I have this series of diagrams that I always like to talk about. I’m sure you’ve all seen this one before, which is the blueprint. It’s what everybody says it feels like to build zero to one. It’s really messy at the beginning and confusing. But don’t worry, it’ll all like level out and you’ll find your path through at the end of the day. And that is absolutely true.
But I think in the era of AI, everybody’s so excited that they also tend to glamorize what this feels like. And so it’s not just this black line that’s messy and it evens out. It’s actually like rainbows and sparkles and colorful. And although it’s really messy, it’s really fun too.
That said, what I’ve realized is that when you’re in that messy part, it can sometimes feel like there’s a bit of a cloud over you because it gets confusing. It gets overwhelming, as I mentioned previously. And I think it’s important to just call that out as a way to give it a name. And then you can move past it. You can understand that, you know, being uncomfortable is natural. It does not mean it’s wrong. And you can really start to just move forward. Bring clarity to chaos. It’s one of the things that I really prize in the folks that I work with—those that can bring that focus to a group and just get them to move forward and really focus on the bigger picture.
Demo 1: Colorizing Old Photos with Imagen (0:06:10 – 0:17:00)
Aakash: So one of the craziest products you guys recently released was Imagen. Can you show us the insider view, the best ways to use Imagen?
Jaclyn: Absolutely. I actually have a side project going on at the moment which is 99 things to Imagen about because I found that the more time I spend playing around with this model, the more I just discover what it’s capable of doing and it makes you think in different ways.
This one I love. You can take any sketch and actually transform it into an art piece now. I don’t have the skills to make a beautiful watercolor blotchy art piece in like 10 seconds. Turns out Imagen does.
What’s also really cool about this is its ability to understand world, like the world model that’s underneath it. I simply asked it to show me what each of these images would look like in winter. I’m from Toronto, so that first image there—in winter, in Toronto, there’s a lot of snow. The Painted Ladies in San Francisco, however, do not get snow in winter. And so the model is not only able to edit the image, it’s able to actually infer what it should look like in that season. That was one of those moments where you just start to realize the possibilities as all of these multimodal capabilities come together.
Aakash: I want to see how to prompt it correctly.
Jaclyn: Okay, this is a picture of my grandparents actually on their wedding day. I’m going to take this image right here and I am going to put it into a chat with Imagen. And then this is a prompt that I actually spent a little bit of time figuring out. And that’s one of the things I would say is when you’re playing around with these models, if it doesn’t work out the first time, keep playing around with things and adjusting it until you get it just right.
Aakash: Can you break down the prompt for us?
Jaclyn: Yeah. So, this one I ended up—you’ll notice like I actually used Gemini Pro to help me figure out what the prompts are. So that’s a good kind of trick is if the image doesn’t turn out exactly the way you want, copy that image, copy your original prompt, put it into Gemini and just ask it like how would you adjust this prompt knowing that the output didn’t quite turn out the way I wanted in these specific ways.
So in this case, I talked about how I wanted vibrant saturated colors. Then it focuses on the lighting transformation of the photo. Then it makes sure to continue to lean into that hyperrealistic detail and texture. And then lastly, this one has a play around with using modern camera and lens optics. When going from old photos to new photos, you want it to feel like it was taken from a new photo.
And then in this case, I actually did end up having some negative prompts that Gemini helped me come up with based off of a lot of the things that weren’t working out the first several times that I was iterating on this. As you can see here, this is the fully colorized version of my grandparents’ wedding photo.
Aakash: That’s wild. Oh man, this is amazing. And one thing I forgot to ask, some people actually have trouble accessing Imagen. How did you access Imagen?
Jaclyn: So I use Imagen in two different places. The first is in AI Studio directly. I have a lot of fun iterating on prompts this way, but then I’ll also use it directly from the Gemini app as well. And it really just is a matter of which one I happen to be in for my workflow. Both of them are easily available. I will also say that we launched Mixboard last week and that’s an open-ended canvas which allows you to also play around with image editing.
Demo 2: Pet to Drone Show to Video (0:19:08 – 0:20:17)
Aakash: This is incredible. What else can we do with Imagen?
Jaclyn: Let me show you another really fun workflow. This is going to be a really long prompt that I spent a while working on. I was trying out how to take pictures of my pets and my kids and turn them into images reimagined as drone shows.
Aakash: Before we see the results, can we look at the prompt for this as well, just to get some learnings on how that was structured?
Jaclyn: Absolutely. So once again, this is one of those examples where I went back and forth in the Gemini app actually to help me iterate and refine. And so the final prompt ended up coming out with your core philosophy on inspiring atmospheric photography. This one ended up really wanting to lean into non-negotiables and then also visual language of drone formation.
We have the universal workflow. This was the part I had a lot of time iterating on, which is “interpret, don’t copy.” It was really difficult to say, don’t literally just take this image and put it in a sky with some drones on top of it. Like reimagine what it is.
And I think just learning how to prompt these models is a skill that takes time, but is incredibly worth it. And that is something that I would also encourage people to do is like spend time just trying to figure out how to coax the magic out of these models because it’s there. But giving a one-sentence prompt is not always going to lead to massive success.
Aakash: Garbage in, garbage out pretty much for these models.
Jaclyn: Yes, very much so. They can develop really specific styles. You can do crazy stuff like we just saw where they take a picture and they reimagine the shape in a figure formation of a drone show, but you really have to iterate on it in this way.
Aakash: So where do we go from here with this drone show image?
Jaclyn: Okay, so in this case, this is how it turned out in AI Studio for this one. I’ll now go to generate media and go to our Veo model. And this is my like fun hack for you all. I’m going to paste this image directly. Upload the image here. And now what you can also do is take it a step further and say “take this drone show image and turn it into a video where the drones fly away to the next formation.” And so it’s going to take a few minutes. But what you’ll end up seeing is the dog literally coming to life as a drone show. And in one version where I did it, my dog’s tail started wagging before all the drones flew away.
Demo 3: Building AI Apps with Opal (0:20:17 – 0:28:41)
Aakash: If you want to create a true ad or something like that, what tools do you guys have to build workflows?
Jaclyn: Great question. So one of the other projects that we recently launched that’s really exciting to play around with is Opal. Opal is—how we describe it—build, edit, and share many AI apps using natural language.
So this is one I’ve already made that called itself wild form. It’s actually pretty fun. It’s a great example of an Imagen workflow. You get a photo and then it’s actually going to generate a nature collage based off of that photo and output it. And so you can see if I click in here, this is that advanced prompt that I had worked on for this particular image.
But what’s really cool about this tool is although this is just asking for a photo using Imagen and then outputting it, you can actually chain together much more complicated prompt chains. And within here you can also change the model that you want to call.
Aakash: Can we see a more advanced example?
Jaclyn: Okay, so this is my custom storybook maker app that I made. And in this one, what I ended up doing was I asked—or well, I designed it so that it would ask for a picture of the main character that you wanted. Then it would ask for their name, and then it would just simply ask where does this story take place.
And from there, I actually love this illustration style. So I put this in as an asset, and then that’s what’s referenced in the image of the character. So generate a kids cartoon image of the person who you uploaded in the style of this particular image. And once again, this is using Imagen for this particular piece of it. And then from there, it’ll also generate a story.
So you end up with three different pages that each have a sort of storyline as well as some contents within it. And this is kind of an example of an Opal that is a much more advanced workflow, but also incredibly easy to use because if you’re starting from scratch, you can simply hit create new and just describe what you want to make in natural language and it will figure out the entire Opal flow for you.
Aakash: So we’re basically chaining together prompts that react to the outputs of other prompts to create a workflow and we can leverage different models along the way.
Jaclyn: Exactly that. And along the way you can ask users for input at various points in the system. And then you can change how the output is displayed. You’ll see in this way I’m just displaying a basic like web page type of an output, but we actually allow you to write to docs, you can write to sheets. And we’re adding more and more features and functionality and integrations.
Building in Public: 10 Side Projects Strategy (0:28:41 – 0:29:45)
Aakash: You mentioned you have 10 side projects going right now. Can you explain that philosophy?
Jaclyn: Yeah, so these are the 10 side projects that I have going at the moment. And you’ll notice I’ve shown you some of them already. There’s my Imagen idea set as well as my writing which is where the resume tips went.
The reason I do this though is because it helps me think differently. It helps me think bigger. And so have fun with it also. I have three kids four and under right now. So I do a lot in the having fun with my kids side of things as well. And that really makes me connect dots in a different way.
And once you feel convicted about your idea, then you can go and actually make a production app and deploy it. And I’m a huge advocate of building in public at this point. I think it’s incredibly important to get that signal and that feedback from users as soon as you can.
Framework 1: Anatomy of an Agent (0:29:45 – 0:34:15)
Jaclyn: Instead of telling you the tactical parts of it, I do want to spend a bit more time thinking about the frameworks that I found helpful because as a product person, that’s usually where I try to spend a lot of my time just to orient myself and understand what makes sense to build. Here are a few frameworks.
The first is just having an understanding high level of what is the anatomy of an agent. Agents have many different components, but at its core, there’s a few pieces that tend to stand out to me.
The first is what are the AI models that you want to use. Do you need to have support for audio, for text, for image? And that’s both image out, but also image understanding. Does it need to be able to write code or produce code? Does it need to be able to understand video or produce video? Just start to understand what are the capabilities you want in your agent or your product and that’ll help give you a sense for which models start to make sense to play around with.
The next is the tools. Models are super capable, but they’re even more powerful when you combine them with tool use. So that’s where you get into things like, hey, should you be calling a search API or should you be using UI actions? One of the projects I’ve worked on is Project Mariner which is an agent that can browse the internet. So obviously that leaned heavily into UI actions and was really trying to push the frontier of what was possible there. And this is also where MCP and APIs come in.
And then another big chunk of it is just how do you think about memory and this is both memory and personalization. What do you want your agent to be able to remember? How do you think about if it should actually be able to personalize an experience or recall things that a user may have previously done? And I think there’s a lot of different ways to build memory, but I usually first start to think about what does memory mean to you? What are the goals that the agent is trying to achieve? What does success for your user look like?
Framework 2: User Interaction Spectrum (0:34:15 – 0:36:35)
Jaclyn: All right. The next one, I like to call this the user interaction spectrum. And I map things out on a scale of, you know, do it for me versus do it with me.
So what do I mean by that? Well, on the do it for me side of the spectrum, you have agents that the user expectation is simply to give them a task and the agent will run off and go do it and then return once the task is complete. Some good examples of this are Deep Research. Arguably, maybe the agent asks you one or two clarifying questions upfront, but then after that it just goes and it searches dozens of different websites and it pulls together a fully fleshed out report for you.
I think audio overviews is another good example of this where you upload a bunch of sources into NotebookLM, another great tool to try out, and you can actually turn that into an audio overview which is basically two people talking in podcast style about all of the source material that you’ve uploaded. But that also takes several minutes for it to go and come up with that audio overview and it’s going ahead and doing that task for you of creating that.
On the other end of the spectrum, you have do it with me. And these are much more collaborative experiences where a user and an agent are basically working hand-in-hand. I think you know vibe coding is a good example of this where it’s a seamless handoff and transition of a user also expecting AI to help them throughout the entire process.
So as you’re thinking about what the goals are of your product, understand how much involvement do you want the user to have in it and that can help you understand where along the spectrum things should lie and that will change how you design the experience.
Framework 3: The Inverted Triangle (0:36:35 – 0:39:51)
Jaclyn: All right. The third framework goes back to thinking and thinking big. I think it’s so important these days with the pace of how fast AI is advancing to make sure that the ideas you have are actually big enough. Otherwise, you’re going to spend weeks trying to build something only to realize it might be commoditized.
So think really big. Now, the hard part of that is sometimes when you think really big, it could take forever to ship. And that’s why I think this inverted triangle framework can come in handy. We think really big, but then we say, “All right, tactically, how do we make sure that we’re getting something in front of users as soon as possible.”
And there’s sort of three different levers that I found myself coming back to time and time again. The first is start by thinking big, but then just reduce the scope and cut features and get realistic about what is really needed in an MVP.
You saw Opal earlier that I demoed. You saw Mixboard earlier that I demoed. For Opal, I mentioned we’re adding more integrations. We’re really going to lean into how can we do more than just docs and sheets. What would it mean to have calendar, to have email, to have all these other tools available to it. We did not set that as the bar before we were able to launch as an experiment.
The next is the positioning of what you’re launching—beta, experiment. Lean into these labels so that you’re setting user expectation accordingly. If you’re saying you’re launching a polished product, that quality bar is much higher than saying, “Hey, I want to build in public. This is an early version. This is a concept edition.”
And then the last is the people or the audience that you’re exposing it to. If something’s super early on, things that we’ve done are just open it up to a small group of trusted testers. It gets people outside of your team using it.
The Paradigm Shift Question (0:39:51 – 0:48:20)
Aakash: What questions should PMs be asking themselves to make sure they’re working on the right thing?
Jaclyn: All right. I think there’s several questions that you can ask yourself as a check-in to make sure that you’re thinking big enough and building something worth building. Two of the ones that I keep coming back to are what I’ve labeled the paradigm shift question.
And that’s really this idea of, you know, are you just building a faster horse or are you building something new like a car? What’s that fundamental problem your user has? And how could new technology create a 10 times better solution?
Another way of thinking about this or framing this is are you just process improving a current workflow or do you think an entirely new workflow should exist for this thing? And I think a lot of times people just focus on the first because it’s comfortable. They know what the workflow is. You can start to say, “Hey, AI can plug into this one feature here and save you two minutes or 10 minutes.” And there’s some value that can be had for things like that, but I think the real value is going to be the unlock on like what’s the new way things will get done. How do you build a car and not just a faster horse?
And then the second one is what I call the future proofing question, which is always wanting to check in with yourself and make sure that you’re thinking about what happens when the models get better. So how will the next AI model update affect your strategy? Will it commoditize a core feature you’re building or will it unlock a new capability that enables you to do more?
Aakash: Can you give us a real example of this?
Jaclyn: I’ll give you an exact example. So with Mixboard before we launched, we’ve been working on it for a while and we actually spent a bunch of time trying to build image editing capabilities into the product. And what we realized was months and months ago, this is pre-Imagen, it was going to be a never-ending hill climb for image editing capabilities.
And that didn’t make sense. So we kind of went back to the drawing board and cut a bunch of features from the early prototypes that we were working on. And then Imagen came out and you realized that, hey, image editing itself is fundamentally changing now. Now’s the time to rethink it. I don’t need the 10,000 sliders that another image editing platform might have had to have if you’re doing it the traditional way. All of a sudden, I have natural language to edit my image.
And I think that’s the other important thing, like models will get better and sometimes you just need to throw out stuff you’ve done previously because it’s no longer relevant. Don’t hold on to it. It got you to where you are now, which I’m sure there’s a ton that you’ve learned. But be willing to let that go and then build for what’s next.
The 6 Characteristics Google Looks For (0:51:50 – 0:57:47)
Aakash: I think you have one of the world’s best views into what it takes to become an AI product manager. What are you looking for when you’re hiring an AI product manager?
Jaclyn: Well, I’ve been thinking about this a lot because I am actively trying to hire an AI product manager. And I think it comes down to these six core characteristics that seem to really matter.
The first one here is exceptional product taste and user-centric craft. And this is this innate ability to just understand what is a good idea. I think that product taste is so important these days. It is one of the hardest things to find in good PMs.
The next is visionary leadership and systems thinking. Being able to connect dots to project out where you think things are going, to be able to paint a picture of the future in a compelling way. So incredibly important when we think about building AI native products.
The third is this clarity in chaos and empathetic resolve. Being able to lead a team through that is incredibly important. Being able to make them feel heard and comfortable and excited, super important in keeping people motivated to move forward.
The fourth is compelling product storytelling. A lot of times, especially in large companies, people try to rely on data as a way to predict what to do. There isn’t a lot of data when it comes to building the next generation of AI native products at a massive scale.
Full spectrum execution and ownership. It’s interesting. One of the things that has been talked about more and more these days is this idea that role profiles are blending. And you need to be able to just kind of work really collaboratively with a group.
And then this last one, deep AI intuition and applied creativity. The ability to have a lot of really good ideas consistently because more and more an idea could be commoditized in the coming weeks or months. Things are moving really fast. I need people that don’t just have one idea and latch on to it and treat it preciously. It’s the skill of being able to have good ideas that I look for.
Resume Tips & Interview Process (0:57:47 – 1:01:33)
Aakash: So that’s what you look for in a product manager. They need to translate that into a resume. How do you create a great AI PM resume?
Jaclyn: Good question. I will share that Opal at the end of this which should hopefully help add some critiques to folks. But I tried to summarize it. The first is just keeping it short. I think some people feel like they need to put their entire life history on their resume and it can get overwhelmingly long. So keep it succinct. And the best résumés I’ve seen are usually only a page.
The next is show, don’t just tell with specific linked examples. More and more résumés are not just a physical piece of paper that you’re handing me. Give me websites to link out to or show me what it is that you’ve done in a way that can jump off the page.
Don’t use vague buzzword-filled statements. What I need you to do is show me that you’re doing those things, not just repeat back at me what it is that I’m looking for.
Designing it with care and personality. I mentioned how creativity is one of the skills I’m looking for. So if I get a super boring resume that just is plain text wall, that to me doesn’t scream I’m a creative PM.
Help me connect the dots of your unique journey. Think about it as telling a narrative or a story. This is the one-page story of you.
Proofread meticulously and check all your links. Frame your impact with context. When you tell me metrics like 50,000 monthly active users, I don’t know if that’s good or bad because I don’t know what the baseline was before that.
And then the last one, highlight your above and beyond projects. That’s why I have side projects on the go. It’s a way for me to do things outside of my day job and stay on top of things.
Aakash: So that’s the resume. Let’s say you make it past, which is hard to do at Google, but you do. What does the interview process look like?
Jaclyn: The roles I have posted, I screen the candidates myself. So there is an initial screen with the recruiter and then the way I’ve been doing it is I actually have candidates answer, I think it’s 5 different questions and I’ve shared them online. Then I read through all of them and the ones that resonate well, I will flag to the recruiter and she’ll get them scheduled for that 45 minute call with myself. And then if that goes well, there is the full round of, I believe it is four interviews with different characteristics that we are looking for.
18-Month Roadmap to AI PM (1:01:33 – 1:03:38)
Aakash: If you had to put it together into an 18 month roadmap for somebody with PM experience, but not AI experience, and they really wanted to break into a Google or a FAANG or an OpenAI, Anthropic, one of these top AI companies, what would be your roadmap for that person?
Jaclyn: I would say focus on building and that includes both building and creating. So it might not be a full production deployed app, although if you want to do that, great. It could just be a series of Opals or maybe it’s more on the creator side that you’ve decided to lean into like some cool videos that you’ve made with AI and talk through the workflow.
Network would be another big one. Go to different events, meet different people. Really try to learn from others, but also create a name for yourself as well. Get on different social platforms, share what it is that you’re building. Position yourself as somebody who has interesting ideas and share those out with the world for feedback.
There’s courses that exist out there that can also be helpful. Read up. There’s so many great Substacks. There’s so many great podcasts that you can be listening to. So I would say just immerse yourself more than anything. Continue to practice good product management first principles.
And then I think kind of the proof is in the pudding, which is why I say create, build, show what it is that you’ve learned rather than just go and do things behind closed doors. I think that getting things out into the open and having people be able to see what it is you’ve done and how you’ve learned is going to be the best way to showcase your skills in this area going forward.
Aakash: Wow, thousands of dollars dropped in value for free, just like you do all the time with your LinkedIn posts and your Substack. Jaclyn, thank you so much for being on the podcast.
Jaclyn: Thank you so much for having me. This was fun. Bye everyone.
Final Takeaways
This conversation with Jaclyn Konzelmann reveals a fundamental shift in how AI product managers approach building products. By using tools like Imagen, Veo, and Opal before writing specifications, Google’s AI product teams are prototyping and validating ideas in minutes instead of spending weeks on PRDs that may miss critical insights.
The three frameworks—Anatomy of an Agent (Models, Tools, Memory), User Interaction Spectrum (do it FOR me vs WITH me), and the Inverted Triangle (think big, ship fast)—provide a blueprint that other product teams can adopt. The key isn’t just understanding these frameworks; it’s applying them to ask the right questions: Are you building a faster horse or a car? What happens when models get better?
As Jaclyn demonstrates through real examples like Mixboard throwing out months of image editing work, the future isn’t about perfecting one idea—it’s about consistently generating good ideas and being willing to abandon work when better technology arrives. The winners won’t be those who hold onto their roadmaps longest; they’ll be those who prototype fastest, build in public earliest, and adapt quickest to model improvements.
For PMs looking to break into AI product management, the message is clear: Don’t wait for the perfect role. Start building 10 side projects today. Use Imagen to prototype your ideas. Build Opals to demonstrate your thinking. Document everything publicly. The proof isn’t in your resume—it’s in what you’ve shipped.
