Head of Instagram Reveals How the Top 1% Actually Work With AI — Silicon Valley Girl Podcast

Adam Mosseri November 17, 2026 45 MIN
Adam Mosseri, Head of Instagram, interviewed by Marina Mogilko on the Silicon Valley Girl Podcast

About the Guest

Adam Mosseri
Head of Instagram (at the time of recording)

Adam Mosseri is the Head of Instagram, overseeing the product and strategy for the social platform used by 3 billion people monthly. A builder and experimenter at heart, he has developed custom AI systems for both personal and professional use, including tools for email drafting, task management, and meeting organization. He emphasizes the importance of feedback loops, continuous learning, and strategic decision-making in leadership roles.

In this episode of the Silicon Valley Girl Podcast, Marina Mogilko interviews Adam Mosseri, Head of Instagram. Adam Mosseri, Head of Instagram, discusses his personal AI command center built to manage the massive operational demands of running an app used by 3 billion people monthly. He walks through his integrated system called "the desk," which consolidates briefings, task management, meeting history, and follow-up tracking from Slack, email, and messages into a single interface. Rather than using multiple AI agents, Mosseri consolidated everything into one main experience, though it contains different functional components like Roger, which tracks promised deliverables from team members and sends automated reminders for overdue items. Mosseri emphasizes that successful AI implementation requires building feedback loops rather than accepting out-of-the-box model outputs. His email drafting system exemplifies this approach: AI generates draft responses, he manually edits them, and the system then compares drafts against actual sent emails to learn his tone, values, and communication style. This iterative process transformed generic email writing into personalized communication that reflects his voice. He cautions that while AI tools evolve rapidly, effective implementation demands a learner's mindset and constant adaptation as capabilities change month-to-month. Beyond his personal setup, Mosseri discusses Instagram's algorithm, clarifying that views alone don't indicate success and that engagement metrics are more meaningful for reaching new audiences. He explains that Trial Reels were designed for specific use cases beyond what most creators understand, and that effective creators should focus on underused features and genuine audience feedback rather than vanity metrics. Throughout the conversation, Mosseri stresses that using more AI doesn't automatically translate to greater productivity, and that personal discipline and clear decision-making about what not to build remain essential leadership qualities.

Key Takeaways

  • Building feedback loops with AI models transforms generic outputs into personalized tools: Mosseri's email system learns his communication style by comparing AI drafts against actual sent emails, progressively improving tone and accuracy rather than accepting out-of-the-box results.
  • An integrated command center consolidates scattered workflows more effectively than multiple separate AI agents: Mosseri merged all tools—meeting briefings, task lists, email tracking, and follow-up reminders—into one unified interface he calls 'the desk' rather than maintaining four separate agents.
  • Tracking promised deliverables through 'Roger' maintains accountability at scale: The system monitors commitments made in meetings, Slack, and email, then automatically surfaces overdue items so leaders can send friendly reminders without relying on memory.
  • Sustainable AI adoption requires embracing a learner's mindset because model capabilities change month-to-month: Mosseri updates his AI usage patterns regularly as technology evolves, resisting the temptation to find a permanent solution and staying in a constant learning mode.
  • More AI usage does not necessarily correlate with higher productivity: Mosseri warns that quantity of AI interactions matters less than thoughtful implementation of feedback loops and intentional decision-making about which tasks deserve automation versus strategic focus.

Adam Mosseri: These models are smart, but they're the average of the internet out of the box. But if you can build loops to help train them, they get much more interesting.

Marina Mogilko: This is Adam Mouseri, head of Instagram, the app 3 billion people use every month. He built his own AI command center to help him run it. Every night, it goes through his meetings, emails, and messages. But the first version, it was garbage.

Adam Mosseri: It was more stuff to read, and it helped me zero. So that loop turned these briefings from a time waster into something that's very helpful if you want to make AI work for you. You have to embrace a learner's mindset. I'm learning every day now in a way that I haven't had to in years.

Marina Mogilko: I'm just thinking like if I'm not a technical person to build out something like that.

Adam Mosseri: I think you need to.

Marina Mogilko: I think you're the top, not even 1%, top 0.001% of people using AI to manage a huge organization. I heard you say you talk to your four AI agents almost every 10 minutes.

Adam Mosseri: The setup itself evolves a lot. I think one of the key things right now if you want to make technology, and in this case AI, work for you is you have to keep a very learner's mindset because what the models and the tools can do today is going to be very different than what they can do in 6 months. And so the things that I use to try and do a better job change month over month because the technology changes. And so it's not like a thing where you can be like, "All right, cool. I'm going to learn how to do this new thing and I'm going to do this new thing forever more." You have to embrace an approach of constantly evolving, which can be exhausting at times, but also very exciting.

Marina Mogilko: So, do you still have your four agents?

Adam Mosseri: Yeah, I have stuff at home and I have stuff at work, but at work, I've put everything into one big experience. But there's different pieces. Last time we talked, I told you about a couple of the pieces. I told you about Roger.

Marina Mogilko: Yeah, we talked about multiple apps and I want to dig into them. From what I remember, you have this app that blew my mind. I'm like, I need to have this. It's the app that has access to your Slack, has access to your email, and tracks things that people promised to you and did not deliver.

Adam Mosseri: Yeah, Roger. I call that Roger.

Okay, that's Roger. Talk to me about Roger.

When you run a large team, one thing that happens a lot is you ask for something and people say "yeah" and then they never do it or they don't do it for a very long time. And so, much to my team's chagrin, I have built a thing that keeps track of things I ask for, either in meetings or an email or in Slack or anywhere else. And then I can go back and take a look and say, "Hey, I asked Max for this thing two weeks ago and he never heard back. Do you want to nudge him about it?" And so I am often sending out a bunch of follow-up notes like, "Hey, friendly reminder, you said this thing would come back in a week and it's been a week and a half. What's going on?" So Roger is one way I try to stay on top of things I ask for and keep accountability high. The way it works mostly is like now it all just gets fed into one big to-do list for me. The way I always pictured things, forget about AI for a second, is like I feel like I have a lot of work to do and there's the thing I'm trying to do right now and there's a list of other things I need to do and that list is so long it goes out of my view. And so I'm like, "All right, cool. I can see like two or three in my peripheral vision of what's next and then that's the one thing I'm focused on." And I actually built something that looks like that where it's here's the thing that I think is the most important thing to do and then here are all the other things and then I can blast through them and I can snooze them or I can edit them or whatever. It's a lot of read this and send back feedback on this roadmap or respond to this question or unblock this team or nudge this person about this thing you asked for two weeks ago. But the thing I will call it like a desk. The desk I built is basically how I've been visualizing my work forever, but now I can actually make it a thing. It's not perfect, but it certainly helps me be more productive.

Marina Mogilko: So desk is like your command center, right?

Adam Mosseri: Where you see all the tasks.

Yeah. Everything's in there. So Caddy is in there, Roger's in there. All my briefings for all my meetings are in there. All the history of all my meetings are in there. Everything is in that one command center.

Marina Mogilko: How do you build it?

Adam Mosseri: We have an internal harness we use for building things and I can use multiple different models. It has basically what we call the meta CLI. So it has a set of internal APIs for things that are inside the company only. But I just built a web app with a simple SQLite database. It's not a complicated thing. But then there are these automations that run at night that try and find all the information it can about how meetings went, what the action items were, what I asked for, what people asked me for, what people messaged me about, what people emailed me about, and then constantly update and populate it.

Marina Mogilko: I want to understand this command center. I'm just thinking like if I'm not a technical person to build out something like that. So as far as I understand it, I use whatever AI I'm using. I'm connecting my Slack, my email, everything and ask it to build a command center. Is there anything else in terms of prompting that I need to think about?

Adam Mosseri: I think we're not at a place yet where if you have no experience building software, you can ask for something like that and get something good. I think we will eventually get there, but right now I think you need to give feedback like, "Hey, this is how I think it should look. Here's what my goals are." It will take a pass. You ask, you give it feedback, it iterates, etc. But I think for most people who have a lot less to-dos than I do, I think something more simple would work great. I think working with any of these proactive agents, obviously ours that we're really excited about is called Muse, to just engage with you via text might be a simpler interface and just be like, "Hey, I want to stay on top of everything. Make sure I don't go more than a day or two without replying to any emails you think are important." And if you give it access to your email, your agent will then just tell you, "Hey, this person is waiting on you. It's been two days. You might want to follow up." You can build much lighter weight things than my overly complicated setup.

Marina Mogilko: What about your emails? I think the last time we talked, 70% of your emails were drafted by AI.

Adam Mosseri: They get drafted, but then I edit them. So here's the thing that I think a lot of people, whether you're really technical or completely nontechnical, I think it's important to understand is these models are smart but they're the average of the internet out of the box and they might not do what you want how you want right away but if you can build loops to help train them they get much more interesting.

You build a loop?

A simple one would be like this email example you brought up. So I have one that will just draft responses to certain emails for me. But I don't have it send those. I take a look at those responses. They're in my drafts folder. Then I edit them and then I send them. But the loop comes into place. I've also told in this case Muse to look at the draft it wrote, look at the actual email I sent and compare them and learn from the delta. And so at first it was writing very generic emails that didn't really sound like me and I was rewriting them and it slowly learned my tone, my values, things like my schedule. And it got a lot smarter over time because it automatically looks at the end of the day every day at all the emails it drafted, all the ones I actually sent and tries to learn from the differences. That's when it gets interesting because if you just ask for it out of the box, you're going to get the most vanilla version. And you want in this case to save time and it's not really saving me time if you're writing emails like somebody else that I would never send. But you don't want to have to tell it feedback manually every time. It's nice to create a loop and then have it learn on its own. So you sort of set it and forget it.

Marina Mogilko: What other loops have you created?

Adam Mosseri: There's a lot. At work for meetings, I have a tool that writes me a briefing. So in a job like mine, you're in a lot of meetings. You get a lot of these really long documents, slideshows or Google Docs or whatever and they can be 10 to 100 pages long. And so I have an agent that basically reads it for me and then tries to summarize it and tell me what matters most. Like, don't miss these three or four things. I still read the full document but it helps me to anchor before I read on knowing what I care about. What are the three or four most important things? I call that a briefing. When I started it, I just wrote a really long prompt to tell the model how to create a briefing for me and it was garbage. It was more stuff to read.

More stuff to read and it helped me zero. But the loop I created was go find the materials after the meeting. So, the PM's email about what the action items are, the AI notes from the meeting if you have those, my notes from the meeting if I took notes in the meeting, any slacks about the meeting if those happened, and then try to grade yourself on how good the briefing was. So, if you were like, these are the three most important things, and we talked about none of those. It's like, oh, Adam actually, this is what Adam actually cares about, or this is what the strategy actually is, or this is what the big contentious decision might be. And so that loop turned these briefings from honestly a time waster into something that's very helpful for me over the course of a month or two.

And it just gets smarter and smarter every week.

Marina Mogilko: That sounds amazing. The question is, do you have to build apps for all these particular use cases or can you just do it within the chat app? Why would you build an app for all the things that you're doing?

Adam Mosseri: You don't need to. For me, it's helpful because this is so much information. So, I'll pull up my desk and I can see all my meetings today, all my meetings for the rest of the week, where I've got the materials have already been sent, where they haven't. For me, that's helpful because I can just be like, "Okay, tomorrow I know is a heavy day, but I've got some time in the morning to sort my thoughts, and Thursday is light or whatever it might be." So that helps me to be able to visualize that. But for just the brass tacks, I'm going into a meeting or I'm meeting somebody. I need context. That could just be a message. That could just be a text.

Marina Mogilko: I wanted to go back to your personal setup that you mentioned, the app that you created for kids with the podcasts.

Adam Mosseri: On different topics.

Marina Mogilko: So I use Notebook LM for podcasts. I think it's very good like the voices are not for kids. I use it to prep for my own thing.

What are the prompts and what are the things that you injected in this app to make it exciting for kids?

Adam Mosseri: So my I've got three boys and they are very into a variety of topics, but one is really into sports scores. So I just built a series of APIs to get really good access to the most recent scores from any of the major leagues and also really grounded in time because LLM struggles sometimes knowing what day it is. So I really built some scaffolding around that idea. And so the main one for Blaze, my middle boy, is usually just like what are the top scores for what are the major games of the Premier League over the weekend, what are the scores and what happened. My oldest is much more into history in certain topics including sports. So he loves things like 10 minutes on the best wingers of all time in soccer or something like that or who are the five best goalies today and why. I built a little app which allows you to just say how old the audience is, how long you want it to be and then what you want it to be about. It's very lightweight and simple. But I picked a voice that was very kid-friendly. I set it up in a way that it would be very engaging for my kids specifically because I know them. I was talking to my dad about it and he's like, I don't understand how AI works. So I just dragged the age slider to 73 and I said explain how AI works. So when you create, you can say what is it? What do you want it to cover? How long is the pod? You can drag it. You can just say I wanted a 26-minute podcast. There's a listener age. My boys are 8 and 10 right now. This is actually indexed to their birthday. So these will change as they get older, but you could just be like 75. You write the prompt and then you hit make and then you get a series of these podcasts. And you've got some basic settings, what model are you using here? You can see I'm using Muse Spark and you set the speed. You can download it. You can send it to friends. So I could send you an email with a file of one. But this is I've built a lot of small apps. This is one of those apps.

Marina Mogilko: Is there anything else you can show?

Adam Mosseri: This one is a trainer. So this is my workouts every week asking about how many things I haven't done anything this week so that's embarrassing but you can see the overall plan all this kind of stuff.

Marina Mogilko: Oh my goodness, that's a sophisticated app. What about you? Can I see your knowledge database?

Adam Mosseri: Yeah. So this one's called Custodian. I switch tools all the time. So we have our own tools, Muse Meta I. I've used Claude a lot. I've tried Duet recently. I've used Typeform a lot. Tried Dots. Used OpenClaw for a while. And when you switch all the time, it's annoying to have to try and re-attach every new tool with all the things about you. So I wire each of them into this thing called Custodian which just has shared activity and memory and skills so that they all can benefit from each other. So if I write like a you know this one's like an email voice. So this will say how to write an email like me and who has access to it. But then they all can benefit from that.

Marina Mogilko: So it connects to all the apps that use all wired in a Custodian.

And every app is instructed to send facts and then it checks whether it already has this information.

Adam Mosseri: So basically there's read and write. So there's a memory system which is just memory as text files. It's all mapped into an embedding space. So you can search based on concept not just on keyword. And so they can all look up memories and they're all instructed to with certain guidelines save memories into Custodian. That way if something else comes out I can just plug it in and see how it works for a while and then unplug it.

Marina Mogilko: But you also, you only use it for your personal. You don't have anything like that for work?

Adam Mosseri: I have it kind of for work, but work is a different system.

Marina Mogilko: How do you use AI to make decisions?

Adam Mosseri: I use it to stress test them often. So if I'm making a decision, I try not to make a ton of decisions. I probably make more than I should.

Marina Mogilko: You have this decision counter.

Adam Mosseri: Yeah. I used to have a clicker. I used to click.

When you manage a large team, you want to delegate as many decisions as you can to smart, capable people who are closer to the work. And if you're making all the decisions yourself, you become a bottleneck. I know this is contentious. I know there's this sort of trend right now in the valley to delegate as little as possible. But my philosophy is a little bit different. And so that doesn't mean I don't make decisions. I make decisions all the time. But whenever I can, if there's someone else who's better qualified to make that decision, I try to delegate the decision to them, but in a very public way so the whole team knows this is actually Max's decision or this is Jenny's decision or you know this is Nat's decision. Because if the team knows it's their call, they act very differently because they know that person's on the hook and that accountability is actually very empowering for those people. Otherwise, the team will often just keep defaulting to me and I don't want to become the bottleneck for progress.

Marina Mogilko: Talk to me about how you're hiring now. Is it different from five years ago?

Adam Mosseri: It's changing really rapidly. It depends on the function. So some of the things stay the same. For all functions, what we use the word function for—you know, engineering is a function, design's a function, data science is a function—so for any hire for a long time I've always looked for a combination of three things: do you have some grit or hustle, can you pick things up quickly, and are you self-aware so that if you've got drive, you're a quick learner and you can take feedback, you can learn how to do anything with enough time? And if you are missing any of those things, you're often stuck. So I've always looked for those three things. But now some of that matters even more, particularly the quick learner thing because things are changing so quickly.

Marina Mogilko: You test for that during an interview?

Adam Mosseri: I think you try to ask questions about things that they wouldn't know about and then see how quickly they can pick it up.

If you have a conversation about it.

You can also take a look at their history. How quickly did they grow their responsibilities in their previous roles? Did they switch types of roles multiple times and find themselves successful? That's another indication that they're learning.

Marina Mogilko: Is that a good signal? Changing roles?

Adam Mosseri: Depends on how often. If you change roles every 3 months, it raises basic questions about why. But if I'm hiring for a very senior person and they spent a long time in data science and then they moved into product staff or in engineering and then they moved into another function and they did a couple years in each, that looks like a pretty well-rounded person to me. And I think that's pretty appealing. But the thing that's happening with all this technology shift is that the lines between the jobs, the roles themselves are getting blurred. And so that's also becoming much more complicated. You've got designers checking in code and you've got data scientists putting together prototypes and mocks and you've got engineers pulling data and doing analyses.

It's all getting blurred.

Marina Mogilko: In this environment, who do you fire?

Adam Mosseri: Firing is hard.

Marina Mogilko: You can't keep up.

Adam Mosseri: What you're looking for is the inverse. You're looking for people who are not necessarily senior but who are growing, who are becoming stronger and stronger, who are getting more capable month over month.

And so the people who make sense less to keep, if you have to let go of people, I think are people who are struggling to evolve and continue to learn in an environment where we all need to continue to learn. I'm learning every day now in a way that I haven't had to in years. And so you just try to focus on helping everybody understand how the world is shifting. Get comfortable trying new things. Get comfortable failing. I feel like learning new ways of working is like learning a new language. You have to be willing to sound like an idiot until you get corrected before you figure it out.

And I think one barrier to entry for learning a second language for a lot of people is that a lot of people can write relatively well, but they can't speak because they're afraid of sounding dumb. But it feels like that to me. You have to be willing just to try and every once in a while it's just not going to work and that's okay.

Marina Mogilko: Is there any way you measure how AI impacts productivity? I know you're against token maxing.

Adam Mosseri: Token maxing is whatever.

Well, we look at who uses the most tokens and half the time it's really productive and half the time it's a complete waste of money and you just try to decipher the difference. But I don't think it's a goal in and of itself.

Marina Mogilko: But how do you know? Do you even care if AI is making someone more productive or is it just the amount of work that they're able to do and you don't care how they use AI?

Adam Mosseri: I care about how productive people are and if AI or any other technology is an opportunity for people to be more productive. I care about them taking advantage of that opportunity.

But you have to be really careful. Large companies can be very focused on metrics. They conflate metrics with goals all the time. A goal might be we want people to feel comfortable sharing more on the platform. The metric might be how many people today share a story on Instagram. But the goal is the metric is a proxy for the goal. It's not the goal in of itself. The goal has to be expressed in English or in a language in prose.

And then you pick the best proxy metric you can. I don't set goals explicitly for AI usage or token usage or diffs or lines of code or experiments per developer, but I look at those things as leading indicators of productivity. But at the end of the day, I care about actual impact. How much did you ship? How much more useful is Instagram? How much better do people feel about Instagram? How many more people use Instagram? Not these AI productivity metrics.

Marina Mogilko: If you've been hearing about AI agents and wondering how you'd actually set up one, I put together a guide that walks you through your first one in about 20 minutes. It helps you choose what to automate and get it working with no coding required. You can get it through my free newsletter. The link is below.

It's such an interesting topic that you mentioned trying to quantify things that are hard to quantify. How do you look at it in general? I don't know, how happy are people with Instagram? Is there a metric like that?

For sure.

And how do you measure it?

Adam Mosseri: We measure sentiment by asking thousands of people, if not more, every day questions on the platform about satisfaction. How fun is Instagram? How satisfied are you with the experience? How much does Instagram care about its users? These types of metrics are called sentiment metrics. Most industries have what's called net promotion score. NPS.

We look at that. We look at how many people use Instagram in a given day. We look at how many people sent a message, how many people shared a story, how many impressions, so how many things someone saw. All these metrics indicate overall the health of the app. But if you overfocus on any one, you risk becoming myopic and doing something that doesn't make sense.

Marina Mogilko: Is there something you still can't quantify?

Adam Mosseri: There's lots that we can't quantify.

Marina Mogilko: What do you do then? Can you give me an example?

Adam Mosseri: Ranking is a great example. There's going to be some angry comments on whatever you post, so just brace yourself. But we rank because there's way more content on Instagram than anyone has time to see, even from the accounts they follow, but certainly from the accounts that they don't but that they might love. And so we want to get them the best stuff. We make mistakes. Ranking isn't perfect by any means. But if we don't rank content, we see not only does usage of Instagram go way down, but sentiment goes way down. Satisfaction goes down. But ranking is imperfect in a bunch of ways, and one of them is you rank for things you can measure, and I haven't seen anything that suggests that there's a correlation between how easy it is to measure something and how valuable it actually is.

So if you think about people's preferences and interests, there's first order preferences and second order preferences. I like how this doughnut tastes is a first order preference. It's a positive experience in the moment. A second order preference is I want to be healthy. I want to eat well and I want to lose weight or I want to gain muscle or whatever it might be. Those two things can be in tension and first order preferences because they are so quick are often easier to measure. Did I like the post? Did I send it to a friend? Am I going to click on the image? These are much shorter interactions. They're much more correlated with your first order preferences and they're much easier to measure than was this really worth my time. Is this something I'm going to talk about with my wife or my best friend tonight? Is this something I'm going to think about tomorrow?

But those aren't any less important. They're just more difficult to measure. And so we are always trying to evolve and try to think of ways to better understand these harder to measure but possibly and often more important proxies for how interested you might be in something.

Marina Mogilko: That's such a great thing because I just did a questionnaire for my audience. I posted three amazing guests and two of them got incredible views and another one I knew was iconic, Brian Chesky. He got less views than average. But I knew from my conversation with him that I've learned so much and we're optimizing for views, but there's something else I'm sure people are remembering. So we did a survey and 44% pointed at Brian as being the best episode.

And it's for me as a creator, I want to optimize for that. I want to optimize for impact, not just views. Because views, you go into this virality strategy and it doesn't bring you satisfaction. I don't get satisfaction from just views. How would you solve the problem with the algorithm?

Adam Mosseri: On your point about your interviews, I think it's super important to say that reach is a means to an end, not an end in of itself. You want reach so that you can evangelize a cause or sell a product or build a brand or nurture a community. And not all views are created equal. The Brian interview reached less people, but it had more emotional impact where the people it did reach, which might mean a more loyal audience or a more excited audience for you long term in a way that's very difficult to measure. Views is a relatively shallow measure of engagement. Did you view it for 3 seconds? That's not a huge bar.

You can look at metrics that might have looked different. I don't know. But if you can look at watch time for instance, that's pretty much the same.

That might be the same thing. So then it might have just been a more narrow audience that cares more.

But that might be okay because maybe that's the brand you want to build or maybe that's the audience that's going to mean the most valuable to your advertisers. There's lots of examples of this. Acquired is a great example. David and Ben are doing amazing things. They do not have the biggest audience. They have a big audience but it's not the biggest podcast audience by a long shot. It is a disproportionately valuable audience.

The audience are people who are mostly or a lot of suite executives who are making pretty big purchasing decisions. And so for their advertisers who are often B2B enterprise products, it doesn't really matter if they got 10,000 or 100,000 or a million views on a thing. As one of their advertisers, you're buying access to a disproportionately important, powerful decision-making audience, even if the number isn't that big. Where one incremental sale might make the entire investment worthwhile.

So it's important to know what kind of business are you, what kind of creator are you, and then not to just lose context in the averages or in the totals.

Marina Mogilko: It is so hard though.

It's very hard because you are distracted by all the signals around you, by all the new people emerging. How do you make those decisions about what not to build at Instagram when there's so much noise? I know you're building a lot, but I feel like we're building too much.

Well, how do you make a decision about not doing something?

Adam Mosseri: I've always wanted to and I have been more and less successful at different times in my career at having not just me but my team focus on doing less things and doing them at a higher quality bar than doing more things and spreading ourselves too thin. In a world where it's becoming so much easier to build more things and there is so much competition and people are becoming more efficient and I want to make sure I'm employing as many people as possible, there's really strong incentives just to do more and more things, and that is just a very tough tension to wrestle with. Our ability to build more things isn't just based on the number of people we have. It's often based on talent density, for lack of a better term. So how many people do you have that you can really trust to make sure that something goes really well? If you have 20 really difficult, gnarly problems and you have a thousand people working for you, but you only have 15 that you can really trust to be fully focused on a really gnarly problem and have it make sure go well, you're in trouble even though you have a thousand people. So we try to look at what is the opportunity, not just today, but in a few years time. We try to invest in things that are going to be more important in a couple years, not less. We try to invest in things that are directly related to our identity. We're about connecting people over creativity. So we're not building a shoe manufacturing plant or something. It has to be related to what we do and it has to be relevant with our most important audiences, which are young people and creators.

Marina Mogilko: You mentioned something about how you think about what's going to be important in two years. For people who are listening, what would be your advice to them based on what you see

Adam Mosseri: as a creator or as an engineer or anything?

Marina Mogilko: Anyone who's looking for a new opportunity?

Adam Mosseri: I think the generic advice, maybe this isn't that interesting so you should call me out if this is a cop out of an answer, is that trajectories matter more than points. It doesn't matter that much where something is today. How many people use a specific thing or how strong an engineer is today. It matters how are they getting better or worse or is the market growing or shrinking and if so how quickly. I would much rather build a product for a market that is growing rapidly than build a product for a bigger market that is declining. I'd much rather hire someone who is relatively inexperienced and not as capable as another candidate but is getting much better really quickly than hire someone who's completely capped out and is no longer really evolving or learning. So whatever your industry is or whatever your job is, I think it's important to have a point of view if you want to think long term about not just where everything is today, but what the trajectories of all those relevant pieces of information might be.

Marina Mogilko: Something that I've talked about to my community as well because I was building this study abroad booking platform, and then study abroad was completely ruined by COVID and I was swimming against the stream. And then I pivoted to being a creator and creator economy is just growing and booming and with less effort I'm able to achieve more. How do you think about the future? What signals are you looking at?

Adam Mosseri: We try not to look at just where is attention but what's culturally relevant, what is resonating with the audiences that we care about most. You can take a look at all sorts of ways to get signals. You can do surveys, you can do ethnography studies, you can run research, you can just read your DMs or your comments, which can be a little crazy, but there's some real gems in there. You got to travel. A big thing for me is to get out. It's hard because I got little kids, but I try to get out there in the world and try and meet people who are using the platform in other countries because the vast majority of our users are not here in the US. And ground yourself in what people's actual experiences are. It's not something you can complete. It's a process that is ongoing.

Marina Mogilko: Before we keep going, I'd love to get to know you a little better. I share a lot about my work here and I'm curious about yours. What do you do and what you'd like more help with? We've put together a little survey. Your answers will help me decide what to cover in future conversations and what resources to put together for you. I'll leave the link in the description. Please take a couple minutes and tell me about you, about your business, about your ambition so I can make better content for you. I really appreciate you. You mentioned something about comments and feedback. How do you work when you're getting so much different feedback from really negative to really positive?

Adam Mosseri: So if you work on a platform like ours, a lot of people are going to use it and even if it's a small percentage of people, there's going to be a lot of strong opinions because it's just a lot of people. You kind of have to embrace that and understand that even though you might get brigaded by a bunch of people who are very frustrated with you or the platform or the team for some reason, it's actually a privilege in a lot of ways to work on something that reaches so many people. So I try to embrace that and then I try to find the signal and I try to ignore the noise. Just because I'm in the job I'm in, every time I post anything, there's going to be a lot of negative comments about small things. You're just going to get a lot of that and you just need to accept that.

Marina Mogilko: So what's a signal and what's noise?

Adam Mosseri: I try to just find the more interesting bits. I try to go into my DMs and just find interesting points of view and questions. I do an ask me anything every Friday on stories and I go through, there's usually a couple thousand questions and I try to find some good questions and answer them. This is actually one area where AI has been pretty useful. I think Muse is going to be, if we're clever, really useful for creators because I have it now read all my requests in my inbox and let me know if there are any interesting questions in there or if any notable people have reached out because I've missed big names who have DM'd me before and I just have not known for a year and been really embarrassed. So I try to do both. Every day it sends me a couple every morning. This person messaged you and you probably should write back, and then here are five legitimately interesting questions and then I'll just go answer them. So I'm outsourcing that a little bit right now, but for years I've gone into my requests folder and through my comments and just try to find a couple people to respond to.

Marina Mogilko: I really like using Muse as a creator because it spots signals that I can't spot with just exporting a CSV from my Meta Business Suite. I think one of the signals it gave me that my family lifestyle content doesn't reach new audience but it re-engages my old audience.

Adam Mosseri: That's a really interesting one.

Marina Mogilko: And all the other tools would tell me to just drop that content because I'm not growing with it.

And I had this internal feeling like I need to deliver for my audience and then Muse confirmed that.

What are the top things creators should be asking Muse?

Adam Mosseri: So it's not actually as good as it should be at a bunch of different use cases. I'm interested in a couple areas. I'm interested in leaning on Muse to help me understand and engage with my audience. I'm interested in leaning on Muse to help me understand what's working, what's not about my content strategy. I'm interested in leaning on Muse to understand what trends are happening out there in the world, not just on Instagram, but on other platforms. There's a bunch of different things that I think it could do really well. Right now it's just not as good as it could be at a bunch of those, but it's getting better and quickly. One interesting question you could ask as a creator is who are the most influential people who follow me. I actually think it's surprisingly good with very little work at understanding trends right now. When I asked about what trends are happening on Threads and on Instagram that are relevant to my job, it gives me a really good answer, which I was pleasantly surprised by because we haven't done any custom work for that. Whereas for basic audience management, there's a lot of stuff that's not wired up. So there's a bunch of stuff.

Marina Mogilko: What's the metric you're optimizing your content for?

Adam Mosseri: I don't think I have one metric. I feel like it's my job to explain what we're doing and why. And that might not convince people. Those people will disagree and that's reasonable. My job isn't to convince people necessarily, but it's to at least explain that there's some rationale and most of these decisions are complicated and there almost always are trade-offs. I try to focus on that. In terms of what correlates most with reach, it's what you'd expect. I think a huge thing that matters for unconnected reach, so reaching people who don't follow your account, it's definitely sends per impression. You always have to look at engagement rates, not just totals. So of every thousand people who saw my reel, how many sent it to a friend? That matters a lot for how much it's going to grow reach people who don't follow you. Whereas for people who follow you, watch time and like rate are much more important. But I try to just focus on that core thing. And so I'll do things that I know don't perform as well, but they're more aligned with the mission or the goal. So I do these creator spotlights where I just ask a creator. It's changed how I do them because what used to work has changed over the years. But if I do one, I ask them for a selfie and I just put a carousel a bunch of their content and I say, "Go follow this person. They're cool."

And I love that because what makes Instagram great is all these creative voices on the platform.

Those carousels don't do as well as a product announcement or a photo dump of my personal life or an explainer about how ranking works. But I don't care because they're 100% aligned with what we want. We want to be highlighting creative voices. We want to make sure that we communicate how much we care about creativity. We want to spread the love and give people their flowers. And so it does well enough that I don't need to evolve the format very much, but it doesn't do as well as my other series, so to speak, for lack of a better word.

And so again, that's a good example of you have to remember why are you there? Why are you posting? It's not just to maximize the number of views you get.

Marina Mogilko: Have a couple quick questions about some other features. DM me to get something.

This is the feature that really helps me get a lot of views, but at the same time, people are complaining that the comment section was something interesting and now it's a repetitive keyword and a generic answer. What do you think about those?

Adam Mosseri: It's a little bit of an engagement hack. I don't love it, but it definitely does work.

Marina Mogilko: It does. And the reach is we tested a couple reels. 100% difference.

Adam Mosseri: But is that a good thing though? From a consumer point of view. So it works as a hack for you to get more reach even if they don't care about the comments. Is that real? It gets way more reach, actually more interesting to more people or is it just getting more reach?

Marina Mogilko: It's getting more reach. I'm getting more subscribers to my newsletter. I don't know. It's working for me.

Adam Mosseri: If people are messaging you and having conversation with you, that's pretty cool.

Marina Mogilko: Don't kill it. People are going to kill me if I brought it on.

Adam Mosseri: It's not the first time I've heard about this one.

Please don't.

They also do comment for it's a thing.

Marina Mogilko: Okay. What's the most underused feature on Instagram that will help people grow?

Adam Mosseri: Oh, I was going to answer one that doesn't not going to help you grow. My favorite features on Instagram are not used by a lot of people. My favorite features on Instagram and they're not going to help you grow, but I'll come back to that blend. Do you know what blend is?

No.

No one knows about blend. So in a DM thread with someone, you can start a blend, which is a feed of reels for both of you, and you can see which is for you and which one's for me. And it's a wildly interesting way to get a little insight into what someone's interests are.

It's rad or your algorithm. So if you go to the reals tab or explore, there's a little couple hearts with some sliders. You can tap and you can actually see what topics we think you're interested in. So it's the first version of being able to see your algorithm and then edit it.

Again, doesn't help you grow, but these are some of my favorite features. In terms of helping you grow, there are no major secrets here. I think for those who create videos, trial reels are really helpful.

Marina Mogilko: Can I drop a feature request here? Trial reels. So right now what I'm noticing on Instagram, you post one video that is maybe you staring the camera, but then you test different overlays with different text. Instagram reads this as the same video.

So I can't really test different.

Adam Mosseri: I think we need to build it in a way where you could just say this is a test and we could treat it as a test. Right now, the point of Trial Reels is not to be a testing platform. It's to be a place where you can post a reel and see how you feel about it and how it does before putting it on your profile forever and bothering your followers with it.

So yes, it's a test, but it's not an AB test. I want to run basically AB testing.

You're hacking it to do AB test. I think we should figure out a way to support that first class so that you don't get caught up. Because the reason why we do that is a lot of growth hackers are trying to hack trial reels just to maximize growth. So they'll post the exact same video with the tiniest tweak 100 times to get 100 AB tests and just to get more reach. And that's obviously an abuse of the system. So it's a balance.

Marina Mogilko: But I think it's the most genius feature. Thank you for it. I think there is so much taste in everything that you do and how you show up. I think it's only you and Branches who have design background in our tech world of S&P 500 companies.

Adam Mosseri: There's not a lot of designers. How has your taste evolved in the last 10 years, especially as you've come into the public eye?

Marina Mogilko: When I build something, even basic stuff like posting a story, what type of music goes with it? Not the most popular trending audio. It has to be intentional.

Adam Mosseri: How do you do intentional? I want to dive into the music one because I have my own little ways I love to do the music stories.

Marina Mogilko: When I was posting about there was Enri Matis's exhibition in MoMA and Enriatis has ties with Russia because there was this Russian merchant who was buying. So I looked up what Chikovsky wrote. It's a lot of details though because that's just for me.

Adam Mosseri: For music for reels and for stories, ideally almost always it has to have some connection to the content. It's usually in the name of the artist, the name of the track or the lyrics in the music. But I like there to always be some through line and then the vibe has to match. I think taste in some ways is a lot just about intentionality. It might not be heavy thinking intentionality or philosophical intentionality. It might just be I want this to be fun.

For everyone who's still figuring out where to start, how has a learner's mindset evolved with AI?

Well, I think it's so easy to become comfortable as you get older in your ways and it can be exhausting to try to learn new things. I think AI has increased the value of being willing to try and learn new things but significantly, as well as increased the demand. So I think it's good, but I think it's a lot. I think it's basically just upped the stakes on how much if you want to stay relevant in a lot of worlds, particularly in tech, you need to learn. But in some ways, for me, it's been very clarifying because I think learning is a muscle in and of itself. You can learn to learn. That's actually really what school's about at young ages. I feel it's forced me to experiment a lot more and try a lot more things. And I've actually as a result gotten quicker at picking up new things and so it's become a positive loop. But it was a real head scratcher for a while.

Marina Mogilko: Thank you so much. I wish we had three hours.

Thanks for having me.

If AI makes it easier to build things, how do you decide what's actually worth building? Watch my conversation with Brian Chesky. We talk about how he comes up with ideas and thinks about what people need. Click here to watch.