Adam Mosseri on Taste, Judgment, and Keeping People at the Center of an AI-Saturated Instagram
Lenny's Podcast
Adam Mosseri has run Instagram for eight years, since taking over from founders Kevin Systrom and Mike Krieger. Before that he designed and led the early Facebook News Feed and ran the team that built Facebook's ranking algorithm. He started as a designer, became a product manager, and now leads a platform that over three billion people use each month. In this conversation with Lenny Rachitsky, he describes how AI is reshaping product teams inside Meta, which human skills he thinks will keep their value, what Instagram's algorithm does and does not understand about its users, and why he expects the flood of synthetic content to help Instagram rather than hurt it. His position throughout is that the people who do well will be clear-eyed about what AI can and cannot do, and that nearly every contested decision about the platform involves real trade-offs.
From a "baker's dozen" to pods of six or seven
Mosseri says the standard product team at Meta has "changed a lot this year." For a long time, a typical team at a company of Meta's size had two or three Android engineers, two or three iOS engineers, two or three server engineers, maybe a generalist, a PM, a designer, a data scientist, and a researcher "if you were lucky." That came to roughly a baker's dozen. The structure had a logic: anyone writing code needed someone familiar with the same codebase to review it, and each specialized function needed its own person. He notes that startups have always worked differently.
This year Instagram adopted what it calls pods. A pod has about four to six engineers who lean generalist, plus a role called "product staff," which Mosseri describes as an evolution of the PM. A product staff member does some of the work of a designer, a data scientist, and a researcher, using the latest internal tools. Specialists join based on what the work needs. A project involving pricing strategy might need a senior data scientist. A project that is truly novel as an experience might need a very senior product designer. The core team usually ends up at six or seven people.
He credits the change to two things. One is AI-driven productivity. The other, which he stresses equally, is that small teams "just by virtue of having less people to coordinate" often move faster and decide better, with "a little bit less design by committee."
How functions bleed into each other
Asked which roles shrink, Mosseri says a pod may have no specialists at all: four engineers and one product staff member, with no data scientist, designer, researcher, or content designer. He sees all the functions "starting to bleed into each other," and the whole industry trying to work out what that means.
His example is data science. At a large company, he says, much of a data scientist's work is fairly mechanical, and only part of it is "art and science." A classic data science task is the waterfall analysis. To study people creating Reels, you map each step of the flow, measure where people drop off, and look for opportunities. That kind of basic analysis is now much easier to pull automatically with internal tools, where before a data scientist had to do bespoke work. A product staff member can do it now, which was not possible a year ago. The result is more people with "generalist shapes," with senior or highly creative specialists brought in when the work really needs them: "a phenomenal product designer or just a genius data scientist or researcher."
Lenny compares this to what Fiona Fung, who manages the Claude Code and Cowork teams, said on the podcast. She hires two kinds of people: builders with great taste who can take an idea end to end, and people with deep expertise in a specific domain. Mosseri agrees that taste matters a lot. He also mentions that Boris Cherny, now known for Claude Code, was once a senior IC at Instagram.
Why he is "long" on designers
Mosseri says taste "matters a ton," because "in a world where it's easier to build things, it's more important to make sure that your time is spent figuring out what you should be building in the first place." He knows many designers feel anxious now that generalists, engineers, and product staff all do design work. Still, he calls himself "pretty long" on designers, because they tend to have taste, and taste is much harder to imagine being automated away.
Lenny pushes back a little. AI-built products often have a recognizable house style ("that's a Codex app," "that's a Claude app," Replit, Lovable), so you might expect design to be thriving already. Yet design job numbers look flat. Lenny's guess is that the missing piece is PM-style business understanding: knowing what will grow the business and what success means.
Mosseri's answer is that it will get harder to judge designers by looking at design roles, because the strongest ones won't stay in them. He cites Nate, a senior designer at Instagram who recently moved into product staff. An excellent designer, he says, probably has strong opinions beyond interaction and visual design, including product strategy, the business, and go-to-market. He expects some of the strongest product staff to come from design and data science. These are people who already had influence across functional lines and can now "just jump in" because the lines are "wildly blurred." They'll be generalists on paper but clearly strong in one craft. He admits his own bias, since he started as a Facebook designer and later changed roles.
Lenny likens product staff to the old GM model, where people from different functions could become general managers. Mosseri agrees that this was always somewhat true of PM and is "so much more true now." He calls this "probably the age of the generalist," but still expects an important role for great specialists. He says he has always romanticized the exceptional machine learning engineer, AI researcher, or shoemaker. That has never been his own shape: "I've never been great at anything. I've always just had range."
Anxiety, and why you still need junior specialists
Asked whether people in data science, user research, or design should think about switching roles, Mosseri says "there's anxiety everywhere." In conversations with people at many companies, he hears concern about competition, job displacement, and unforeseen consequences of technology moving this fast.
He expects functional lines to keep blurring but thinks functions will survive in a different form. Not everyone will be a senior IC, but everyone should be either senior or on the way there. His argument: "You can't just have a bunch of super senior data scientists and like no new ones because then who's going to be the new super data scientists in the future." Companies need to hire, mentor, and grow talent. Teams may be smaller overall, and people who aren't on the path to becoming very senior specialists could move into generalist roles, which he calls "a reasonable soft landing." A company that invests only in today's senior talent and not tomorrow's will, he thinks, regret it in a couple of years. He immediately adds a caveat: "who knows what the world looks like in a couple years." His general advice is not to be overconfident in any prediction "because there's just too much flux right now." Lenny notes that Benedict Evans said much the same thing on the show.
What he hires for now
Mosseri has long looked for three traits in any function: grit ("some fire in your belly"), fast learning, and self-awareness, so a person can take feedback and know what they are and aren't good at. With all three, he says, you can eventually get good at almost anything. When one is missing, "there's usually an issue."
For the next five to ten years, he adds two traits he keeps pushing himself toward: staying curious and putting yourself out there. Because no one really knows what's happening, you have to be willing to try things. He compares it to learning a language (Lenny speaks Russian). His guess, which he says isn't based on research, is that the best predictor of learning to speak a language is whether you're willing to sound like an idiot, be corrected, and not take offense. The same goes for new tools and models. People who are curious and try things will adapt. Those who aren't curious or won't risk mistakes are "in a ton of trouble," or at least in for a hard time. He sees these as premium traits across many industries over the next ten or twenty years, not just at Meta.
As for what's declining, he expects fewer roles for leaders of very large organizations, a craft he calls distinct from management. Those jobs won't disappear, but there will be more small teams and fewer people managing thousands, so he'll hire for that less.
AI reshuffles who succeeds
Lenny raises something he has heard from others: AI is resetting who performs well, lifting some people who struggled before. Mosseri agrees and puts it down to the job itself changing. Engineering used to be perhaps 40 to 60 percent writing code. Now, especially at the AI labs, engineers spend most of their time planning and reviewing code. "That is a very different job." Someone who loved writing code might hate it. Someone who was slow at writing code might love it. Who succeeds depends on whose strengths fit what the tools and the business now need.
He also points to people who had good ideas for other functions but lacked the mechanical or technical skills to act on them. AI lowers that barrier. His own case: when Facebook hired him, designers had to be able to program and went through a technical interview loop. Facebook later dropped that requirement because it made hiring too hard. Now he's programming again for the first time in about ten years. "I'm a mediocre engineer on a good day but now I can write code responsibly." He sees the same thing at every level: designers programming, engineers doing solid data analysis, data scientists drafting design proposals.
He adds that the tools "aren't all great." He objects to the polarized question of whether someone is "AI-pilled" or "anti-AI," which he raised with his team the day before. People aren't binary and neither are the tools: they're "amazing at some things and remarkably bad at others." The people who get the most out of them will be clear-eyed about current strengths and weaknesses and will also have "a nose" for what AI will be good or bad at a few months out.
Token spend, cost, and the model curve
Lenny jokes that we used to be able to write code for free. Mosseri says you still will, just with a smaller model. He asks listeners to compare what a given price buys in intelligence today with what it bought a year ago, and predicts that the extra value of frontier models will stop mattering for small programming projects. He says he spent a lot of time that week with Fable and Mythos from Anthropic, and "for the first time I'm like, 'Oh, I'm just talking to a much more technical, much smarter engineer than I am.'" He guesses that a year from now, Fable will probably be "just fine" and fairly cheap for his side projects, with no need to pay for the latest frontier tokens. Lenny recalls Kevin Weil's line from his time as OpenAI's CPO: "This is the worst the model will ever be."
On Meta's reported token-spend leaderboard, Mosseri is blunt: "It's a terrible idea. No leaderboards for token spend." He says costs have been reined in somewhat by shutting down "the silly things" people were doing. "It's not that hard to build a token incinerator," and once you compare dollars in to value out, some uses are plainly bad ideas. For now there are no token limits for engineers, or he thinks for anyone.
He expects limits eventually, especially if costs rise before they fall. He treats tokens like any other resource he allocates across teams: GPUs, CPUs, storage, RAM, opex for labeling budgets, payroll for headcount. He can picture a strong engineer's token burn rate matching their cost of employment within a year or two. At that point caps make sense, set in proportion to the company's trust that the person will spend in an ROI-positive way. His forecast is that costs will rise because usage grows, not because prices rise, and then fall as frontier models get into a price war. "I think it'll be a bit of a roller coaster."
Where human judgment still matters: vision and strategy
Asked where human brains stay most valuable as AI takes over more of product development, Mosseri names taste and judgment, especially on strategy. You might get feedback on a strategy from AI, he says, but you won't ask AI to come up with one "anytime soon," unless it's within bounds you've set: your goal, vision, constraints, budget. He thinks this looks more like management: define success, decide how prescriptive to be about the path, and give feedback along the way. He wonders whether a leadership tension he knows well will carry over to agents. Too prescriptive and you stifle good ideas. Too open-ended and teams waste time heading in the wrong direction.
He defines vision as a description of the world or product state you want to reach, and strategy as "an opinionated path to achieve that vision." A strategy can't just be "be the best." It has to be controversial enough that a reasonable person could disagree, "because otherwise you're probably just trying to compete on raw execution." He expects people to spend more of their cycles on vision and strategy and less on execution.
Lenny asks why AI isn't naturally good at strategy, since it can take in the market, competitors, and metrics. Mosseri says it could be, but in his experience it isn't unless you steer it hard, by constraints rather than toward an answer. Strategy has many inputs: the state of the technology, the team and what motivates them, the fact that an idea "on the bubble" can attract top talent, the competitive landscape, and for a company Meta's size, regulation, compliance, and the brand's reason to exist. Ask lazily and you get "something pretty predictable that probably the competition would expect you to do." Lay out the inputs, make it a back-and-forth conversation, and tell it to be critical, and it can be helpful and clarifying. He recommends choosing a model that pushes back. Lenny says Mythos has gotten good at refusing and moving on. Mosseri says one model has "always been a little bit of a jerk in a way that I actually appreciate," because he wants real intelligence, not a pleaser.
Lenny brings up the centaur and reverse-centaur idea he attributes to Cory Doctorow. A centaur is a human directing a horse's body, and a reverse centaur is a machine directing a human, as with gig workers following app instructions. Lenny suggests nobody wants AI dictating strategy for them to carry out. Mosseri agrees there's a lot to be careful about: just because a workflow can be outsourced to AI doesn't mean it should be. Some cases are win-win, and in others the risks outweigh the benefits.
Great product leaders as curators
Mosseri says that while some excellent product leaders are prolific "idea machines," many of the best are curators "of people, … of ideas, … of technologies, … of strategies." When he hires a lead for an area, he doesn't care whether the strategy comes from them. He cares that the strategy is excellent, that everyone buys in, and that it's executed well. You need some ideas of your own to curate well, he says, but the best leaders accept they can't come up with everything and build an environment where good ideas surface and get chosen.
Curation also means fit. For an area like trust and safety, he has leads in engineering, product staff, data science, design, and research, and he needs the five to complement each other in skills and weaknesses. They also need good chemistry, which he calls "more art than science." "A leadership team with strong trust and rapport can work through most anything. A leadership team without trust or rapport, like anything can become an issue." It works the other way too: he has had two people he thought were excellent, and whom he liked, who simply couldn't get along. "This isn't a competency issue. This is just a personality issue," and sometimes you have to split them up.
What the Instagram algorithm actually knows
Asked what the algorithm knows about human behavior that people might not realize, Mosseri turns the question around. The common misconception is that the algorithm has "a much more detailed semantic understanding of everybody's interests and preferences" than it really does. Over the past five to ten years, most progress in recommender systems has come from large embedding models and similar techniques that produce artifacts people can't read: giant vectors of numbers. When people imagine the algorithm "knows I like surfing," it actually "just has this big-ass number that happens to correlate with surfing."
He says LLMs are starting to change that, because they can describe in plain language what those unreadable artifacts approximately mean. That week he had posted about a feature called "Your Algorithm." Everything you've interacted with sits in an embedding space, which he describes as a map where similar videos sit close together. An LLM can be asked to describe a region of that map, and it might say "deep pour-over coffee snobbery." Concepts can be embedded in the same space as well. The feature lets users see the topics Instagram thinks they're interested in and add or remove them, which he frames as giving people some agency back as social apps are increasingly driven by recommendations.
He describes other controls as future work: non-topical requests such as more fun content, more friends, not seeing high-school acquaintances' kids' photos, or not seeing seven photos in a row. "We have a lot of work to do." Until recently, he says, the system mostly worked like this: you liked these photos, people who liked them also liked those, so you might like those, which he admits is an oversimplification. "Only now are we actually getting as sophisticated as I think people have assumed we've been for many years."
Why chronological feeds disappoint
Lenny notes the recurring pattern: people say they want a chronological feed of accounts they follow, yet engage more with algorithmic feeds. Mosseri says he had posted about agency that week and "just got destroyed in the comments," which he calls part of the job.
He gives two problems with chronological feeds. The first is incentives at scale. If ordering is purely by time, everyone's incentive is to post as much as possible to stay on top. The New York Times can publish 50 things a day while your best friend might post once a week, so professional and publisher content takes over. He compares designing these systems to designing a city: you have to think through the mechanics, the incentives they create, how people will behave, and what follows. The second problem is that the most interesting thing often isn't the newest. Recency matters for relevance but isn't the only factor. His example: if his sister in Germany got engaged overnight, he'd want to see that before a photo of his brother's po'boy sandwich. (Lenny points out his sister is already married; Mosseri says she married last year, which is why it came to mind.)
He wants to give people agency in a way that works for the whole system: for the business, which he says openly matters, and for the community. Instagram has made chronological the default, and he says the results were lower usage and lower overall sentiment. An individual may like the choice at first, but over months of seeing less interesting content, large-scale surveys show people becoming "less and less satisfied with Instagram."
AI content as a tailwind, with conditions
Asked whether the rise of AI-generated content helps or hurts Instagram relative to other platforms, Mosseri says, "I think it's going to be a tailwind, but I think it's going to be a challenge." More content can mean more attention in an attention-based advertising business, but it comes with costs. He admits Instagram isn't yet very good at ranking AI content. Some is great and some is "crap," and users should see only what interests them.
His main argument rests on a view he has held for years: power across industries is moving from institutions to individuals. Sports is his example, where players now matter more than teams, which wasn't true when he was a child. So Instagram should invest in individuals and creators, defined broadly. Not just influencers making branded content, but anyone using the platform to do what they do: a journalist, an artist, someone selling scarves they sew. Creators have been one of Instagram's two or three most important audiences throughout his tenure. "In a world where there's an abundance of synthetic content, I actually think people are going to seek out creativity and authenticity and people more, not less." Instagram, he says, "was never just about the content." It was about the person behind it, their point of view, and why they shared it. He adds that "we are not the best at a lot of things," but by his account, based on how Instagram defines creators, it is the largest creator platform, which is why he expects the tailwind.
That doesn't mean removing AI content. Unsafe content is taken down as always, and interest is handled by ranking and personalization. If you love AI content, "you should be able to have a feed that's just like AI town." He says he understands the concern and isn't oblivious to the paradigm shift, but "I don't think we should judge content based on the tool that made it." He wants to judge content by its substance, point of view, and author: "I don't think we should filter out AI content. I think we should let you know if content is AI content or not." He also wants users to know more about who posted something, such as where they are, how often they've changed their profile, and whether the account is three days or three years old, so they can decide whether to trust it.
He names two AI creators he enjoys. One is Plastic Dream Sequence, which makes short, somewhat strange clips of Barbie-like dolls singing. He likes that it's "very clearly AI," "not pretending not to be," and has "a very clear creative and aesthetic point of view." The other, which he calls "If Only AI," is a creator in France, he believes in Paris. He uses different models for images, video, and music to build animated dreamscapes with a distinct aesthetic, and Mosseri says you could think of him as a painter for whom these are the tools.
Labeling content, labeling accounts, and AI spam
Mosseri separates labeling content from labeling accounts and considers both useful. For content, you should be able to ask "is this AI?" and get an honest answer: probably, not sure, definitely not, or definitely. He calls this hard. Instagram can detect AI content now, but as models improve it "might lose the ability to detect that," so it needs to be candid about its confidence. Long term, he thinks it may be more practical to label camera-captured, non-AI content than to label AI content, "for a couple of reasons."
For accounts, he describes a new spam vector. AI creators are fine in themselves, he says, but some fake accounts, such as an "AI monk" selling bogus supplements without making clear it's AI, exploit an aesthetic or stereotype. Instagram needs to crack down on those. The principle is that users should know whether an account is a real person and whether content is real, and then decide for themselves.
What he admires in TikTok
Asked what other platforms do well, Mosseri names TikTok's recommender and its ability "to break small talent." He explains the difference between exploitation-based ranking, which uses what you already know about a person, and exploration-based ranking, which tests content to find what someone might like without knowing it yet. His example: you like Bieber, so the system tries showing you a genre you've never engaged with to see what happens. Exploitation moves engagement more easily. Exploration is harder but good for niche and small creators, because it helps them find audiences that would otherwise never have seen them.
He says Instagram has spent the past couple of years ranking not only for engagement but also for originality, for the number of posts that break out, and for recency to stay culturally relevant, with much of that inspired by TikTok and ByteDance. By Instagram's best estimates it is ahead in a couple of areas and behind in a couple, and he sees "line of sight" to being best in class at recommendations for the first time in his tenure. When Lenny congratulates him, he says, "we're not there yet." His team calls him "disappointed dad" and has asked him to ease up, so he's trying to give people more credit. Lenny notes that never being satisfied is common among successful leaders, and Mosseri calls it "a blessing and a curse."
Trade-offs and public criticism
Lenny notes that the global algorithmic feed people complain about is also what lets new creators reach large audiences. Mosseri uses this to make his broader point: nearly every contentious issue involves trade-offs. If you never want to see anything uninteresting, you'll get "the most basic general lowest common denominator stuff." If you want discovery, you'll sometimes see misses. The same goes for privacy versus safety, such as whether a company should scan messages. When he argues publicly, he says, he usually isn't trying to change the minds of people who feel strongly. He's trying to lay out the pros and cons for everyone else reading.
Asked why he became the public face of so many controversies, he says it began when he ran News Feed. His view was that the debate would happen with or without Facebook, so Facebook might as well take part. He became active on Twitter, where journalists were at the time, as a gesture of humility to "show up on their turf." Following all the company's biggest critics made his Twitter "the darkest place in my life," which he says is not a knock on Twitter. His reasoning is that Meta's products are a big part of daily life for many people, change brings anxiety and scrutiny, the company has been criticized both fairly and unfairly, and "people are fundamentally more afraid of things that they don't understand" and of companies that are secretive and inaccessible. He also recalls a long-ago debate in Mark Zuckerberg's senior leadership team about why a social media company relied on conventional press releases instead of its own platforms. Mosseri wasn't part of that debate but inserted himself to mediate, and Zuckerberg responded by suggesting he try it and see how it went.
On coping with hate, he goes back to the 2009 News Feed redesign, launched in March 2009 when he was a 25-year-old entry-level IC designer. The first comment the team saw was derogatory, homophobic and antisemitic, and asked, with expletives, who had designed this. It was him, and he was devastated. He came to see it this way: if you spent 30 to 50 minutes a day at your desk organizing photos and writing to friends, and someone rearranged it without warning or explanation, you'd be reasonably angry, and that was happening to millions of people. He tries to keep things in perspective and to step away with his kids and time outdoors. Some months it's easy, and some months "it's really really grinds on me."
The feed-redesign backlash, and testing at scale
Asked about the widely protested move toward a full-screen video feed and its reversal, Mosseri says several things got conflated. The redesign was a test to 4 percent of iOS users, never meant to launch, and intended to gather early feedback. At the same time, Instagram was leaning heavily into Reels and into recommendations from accounts people didn't follow, and some creators blamed ranking changes for falling reach. Prominent creators criticized Instagram publicly, the press covered the backlash, and more creators joined in, which he calls a "multiplier effect" between creators and the press. Instagram has kept investing in video, creative tools, ranking, and recommendations since then, and he says that has driven most of its growth in the years after.
His takeaway isn't that the test was wrong. It's that Instagram could have explained things better and "was just pushing things a little bit too fast," and that for a platform like Instagram you have to be realistic about how quickly you can change it. He'd still take a backlash like that every couple of years over not evolving: without that willingness, he says, Instagram wouldn't have video, DMs, Stories, or ranking.
Because of this, every potentially controversial test now gets a communications plan in advance. The team assumes "not if it leaks, when it leaks" and decides the message and whether to talk proactively or reactively. "You can't launch something to three billion people and not test it first, but you can't test something at our scale and not expect people to cover it," which he says makes the development cycle more complicated than it used to be. Lenny mentions a pricing experiment by Anthropic's head of growth that blew up on Twitter. Mosseri says pricing tests in particular need great care and that companies "should all share notes more."
Two failures: Facebook Home and the first Reels
Mosseri gives two failures. The first was his first project as a PM: Facebook Home, a fork of Android at the operating-system level, plus a piece of hardware made with HTC. "It was a spectacular failure." He was a design manager, and when the project's PM quit, he declared himself PM and threw himself into learning about carriers, OEMs, certification, Android, and operating systems. He says he learned more in that year to year and a half than in probably any other year of his career. He's glad he brought the project to an end, and he argues that sometimes the best thing you can do is execute an idea without market fit well, just to find out whether it was a good idea to begin with.
The second was at Instagram: the first version of Reels, built on top of Stories around 2019, because Stories was growing fastest. It wasn't a strong foundation. Stories has a relatively low read-through rate, since there are far more stories than people have time to watch, so most of those Reels were never seen before they disappeared. Mosseri believes that if Instagram had launched in summer 2019 the version of Reels it actually launched around summer 2020, TikTok would still be big and important but not as big as it is now. By his account, TikTok exploded in 2020, when people were stuck at home during the pandemic, looking for some joy and fine with sound on, and Instagram "was out of position." As a designer, he had wanted to extend existing primitives instead of adding new surfaces, "on the other hand, I was wrong," and he calls it a big fork in the road for the business over eight years. When Lenny jokes that this created economic opportunity by letting TikTok grow, Mosseri says he's glad TikTok exists.
Kids, screens, and vibe coding with a 10-year-old
Mosseri's three children are 10, 8, and 6, which he says is too young for social media. Each has an iPad, and screen time must be earned. For example, three half-hour homework sessions earn 90 minutes on the weekend. The point is that screen time can't just be granted whenever they ask. He approves every app, and he says parents should approve what apps kids download, something Meta has long advocated at the policy level. Plane flights are an exception, which he treats as "just about surviving."
He's also experimenting with his oldest. He supports keeping phones out of classrooms and thinks that's probably good for education. He considers the worry that kids using AI won't learn critical thinking a valid concern, but he's also worried about kids who don't learn to use AI falling behind, and he thinks both need balancing. His son loves video games, so they've been vibe coding together. They built a 19-level platformer that looks like an 8-bit Super Mario, with themed levels, different monsters, and a store for skins and weapons. He calls it "unbelievable" what a 10-year-old "who still types with three fingers" can do in a couple of hours sitting together. The goal, he says, is for his son to make things and think, not just play. He tried this six months earlier and it "totally didn't work," but newer models have made it "amazing." His son is currently using Claude Code. His principles are boundaries, limiting use to activities you consider healthy for your child (each child is different), and raising kids to be digitally and AI literate without making it a free-for-all.
What he asks of listeners
When Lenny asks how listeners can help him, Mosseri returns to his central theme: "this world and technology is complicated and there are almost always tradeoffs." People are free to disagree with decisions he or Instagram make, and he says he welcomes the criticism. He asks them to remember that the people making these decisions are doing their best, and that "none of these contentious debates are nearly as simple as most people pretend to make them out to be."
No, I think taste matters a ton. In a world where it's easier to build things, it's more important to make sure that your time is spent figuring out what you should be building in the first place. The people who I think are going to make the most of it are the ones who are clear-eyed about what AI is good at and what it's not good at and also have an instinct or a nose for what it will be good at and not good at.
What's something that the Instagram algorithm knows about human behavior that people may not realize?
I think people assume that there's a much more detailed semantic understanding of everybody's interests and preferences in the algorithm than there is.
Is the rise of AI content a headwind or a tailwind for Instagram versus other platforms?
I think it's going to be a tailwind, but I think it's going to be a challenge in a world where there's an abundance of synthetic content. I actually think people are going to seek out creativity and authenticity and people. I don't think we should filter out AI content. I think we should let you know if content is AI content or not. That's hard, by the way.
Where do you think human brains will continue to be most valuable as AI continues to eat more and more of that product development life cycle?
That's a great question.
Today my guest is Adam Mosseri, head of Instagram. Over three billion people use Instagram monthly. That's one in every three people alive. It boggles the mind. Prior to Instagram, Adam designed and led the early Facebook News Feed. He also ran the team that built the Facebook ranking algorithm. And 8 years ago, he took over Instagram from its founders, Kevin Systrom and Mike Krieger. He's a designer turned product manager turned leader of Instagram. Adam is also famous for being the face of all of the controversy and changes that come with evolving Instagram as a product, which we talk about.
Before we get into it, don't forget to check out lennisproduct.com for a free year of the most interesting and well-crafted AI products in the world, available exclusively to Lenny's Newsletter subscribers. With that, I bring you Adam Mosseri.
Adam, thank you so much for being here. Welcome to the podcast.
Thank you for having me. Excited to be here.
You've been doing product for a long time. You get to see how a lot of teams operate across Meta within Instagram. What does just kind of like the canonical product team look like in 2026? What's kind of most different today in how teams operate slash should operate versus say a couple years ago?
It's changed a lot this year. So for the longest time at a big company like ours, the canonical team was something like two or three Android engineers, two or three iOS engineers, two or three server engineers, maybe a generalist, a PM, a designer, a data scientist, a researcher if you were lucky, and maybe that's about it. So on the order of a baker's dozen. And that is a function of, you know, you want to have for anybody who's writing code someone who can review their code and who's familiar with that codebase, and having these different functions that are more specialized. You know, I think it's very different at a startup, but this year it's changing. We've adopted what we call pods, which are just mini teams where it's call it four to six engineers who are a bit more generalists. One we call product staff, which is sort of an evolution of the PM. So a PM who can do some of what a designer does and some of what a data scientist does and some of what a researcher does, leveraging the latest tools that we have for them. And then whatever specialist they need: if they're doing something that requires a pricing strategy, you need a senior data scientist. If you're doing something that is really novel from an experience standpoint, you need a very senior product designer.
So we try to build the team based on the needs of the work a bit, but then end up with a much smaller core, which is more on the order of six or seven usually. And that is a very big shift that's just happening to us this year. But just by virtue of having less people to coordinate, they can often move faster and make better decisions, a little bit less design by committee. So we talk a lot about AI adjusting and improving productivity, and that's part of it, but I think another part of it is just the small teams I think often are just more effective.
This episode is brought to you by our season's presenting sponsor, WorkOS. What do OpenAI, Anthropic, Cursor, Vercel, Replit, Sierra, Clay, and hundreds of other winning companies all have in common? They are all powered by WorkOS. If you're building a product for the enterprise, you've felt the pain of integrating single sign-on, SCIM, RBAC, audit logs, and other features required by large companies. WorkOS turns those deal blockers into drop-in APIs with a modern developer platform built specifically for B2B SaaS.
Literally every startup that I'm an investor in that starts to expand upmarket ends up working with WorkOS. And that's because they are the best. Whether you are a seed-stage startup trying to land your first enterprise customer or a unicorn expanding globally, WorkOS is the fastest path to becoming enterprise ready and unblocking growth. It's essentially Stripe for enterprise features. Visit workos.com to get started or just hit up their Slack where they have actual engineers waiting to answer your questions. WorkOS allows you to build faster with delightful APIs, comprehensive docs, and a smooth developer experience. Go to workos.com to make your app enterprise ready today.
I love this. So on this team of six to seven, what's the makeup again? And which role are you finding you have less of, if you're going 50% size?
You just have less specialists, right? So you might not have any. You might be four engineers and a product staff and there's no data scientist, there's no designer, there's no researcher, there's no content designer. The product staff is the generalist that sort of supports all of those things. I mean, what's clearly happening is all the functions are starting to bleed into each other and the whole industry is wrestling with what that means.
You know, a lot of what a data scientist does at a big company, for instance, is relatively mechanical. And so there's stuff that they do that is really more like art and science, and the stuff that's really more like just pulling data, data management. So some of the tools that we're building internally to understand, for instance, a traditional data science question would be a waterfall. So if you wanted to look at people creating Reels, you would look at all the steps and then how people fall off on each step and try to figure out where there might be opportunities to improve things. That kind of basic waterfall analysis is much easier now to use some of our internal tools to just pull automatically, as opposed to having to have a data scientist do a bunch of bespoke work for that. So a product staff might be able to do that now, and they couldn't do that a year ago. So you just end up with these people who have more generalist shapes, and then when you really need it, you have a more senior, ideally, or just more creative specialist. So, you know, a phenomenal product designer or just a genius data scientist or researcher.
This is so interesting. It's exactly what I just heard. I had Fiona Fung, the head of engineering for Claude Code and Cowork, on the podcast. She's Boris Cherny's manager, and she described the people she hires now are, one, builders with great taste that can take an idea from end to end, and people with deep expertise in a very specific domain.
The taste matters a lot. I really agree with that. Boris used to work at Instagram. I love
Yeah. He was a senior IC at Instagram for a while. I love seeing him. He's all over Threads now. It's like he's sort of like the face of Claude Code. He's a celebrity now. He really is. He's his own little... Yeah, for sure. In a world that is like on fire right now.
No, I think taste matters a ton. So in a world where it's easier to build things, it's more important to make sure that your time is spent figuring out what you should be building in the first place. Actually, a lot of designers right now are very anxious about their roles. You know, you've got these other generalists doing design, you've got engineers doing design, product staff doing design, but I'm actually pretty long on design or designers because they tend to have taste, and I think that is something that is much more difficult to imagine being automated away. And so there's other challenges with design sometimes, but I'm pretty long right now on designers.
I've always felt that too, as it is so easy to build, and all the work that AI produces is so like, you can tell this was Claude design. I know what you did here.
This is Codex.
Well, they all have their vibe, right? Like you vibe code your apps. We call it vibe coding. You're all like, "Oh, that's a Codex app." Or, "Oh, that's a Claude app."
Right. And that's Replit. That's Lovable. Like you can
Yeah.
You can predict these things. I've always thought that too, that design should be thriving already now. For some reason, it hasn't yet. If you look at jobs for designers, they're kind of flatlining. I feel like the missing piece is like the PM piece of deeply understanding the business and what will grow it and what success is, like all that stuff, the business side of it versus the taste side of it.
Yeah. I think you're going to see, you know, we have a senior designer at Instagram called Nate who just transferred into product staff. So I think some of what you'll see is it will be harder to talk about design roles and who's a good designer, because they're not going to just stay in traditional design roles. You know, if you're an amazing designer, you probably have strong opinions outside of just the interaction and visual design. You probably have strong opinions on product strategy, even on the business, on the go-to-market. And so I actually think some of our strongest product staff are going to be converts from design and from data science who are just looking to expand their reach. And they were influential across functional boundaries before, but this world where those functional boundaries are just wildly blurred just allows them to jump in. And so sure, they'll be technically a generalist on paper, but they clearly have a uniquely strong ability in one type of craft, but they've got the ability and strong opinions to make informed decisions across other parts or other crafts. And so I don't know that all the strongest designers I have will all be in design. They probably will be the majority, but I can imagine a bunch of really strong ones moving roles. But I should also check my own bias here, though, because I started as a designer at Facebook way back when, and I switched roles.
No, designers are great. I'm a big fan. So, this is really interesting. There's always been this like GM model where different types of functions can become GMs. It's like this product staff role feels like a similar situation where different functions can become product staff.
Yeah. Yeah. And that was true of PM before, but it's just so much more true now. And in some ways it's probably the age of the generalist, but I still think there's going to be a real important role for these really amazing specialists who are just all about going deep. I wish I was like that. I always romanticized the phenomenal machine learning engineer or AI researcher or shoemaker. Like I think that's the coolest thing in the world, but it's never been my shape. I've never been great at anything. I've always just had range. That's always been my strength.
Same. Okay. So this idea of product staff, so the idea is on these new pods. So this is like a new thing you guys are doing. So there's these pod teams, product staff, engineers, and maybe one specialist that's going deep on, say, pricing algorithm or something like that. So what this tells me is there's these adjacent roles that are maybe more in trouble over the years. Data science, for example, user research, for example. You talked about designers being anxious. Is there anything there of just like, oh, maybe folks in these groups should think about shifting to other roles?
I mean, there's anxiety everywhere. I've talked to a lot of people at a lot of other companies, and it just seems like there's a lot of concern right now about competition, about job displacement, about unintended or unforeseen consequences of all this technology and all this moving so quickly. So that's definitely happening. I think that you will see the functional lines continue to blur, but I still think there will be room for functions. They'll just be shaped differently. They won't all be senior ICs necessarily, but they'll all be either senior or on their way to being senior. Like you can't just have a bunch of super senior data scientists and no new ones, because then who's going to be the new super data scientists in the future? So you need to basically hire and mentor and grow talent. You know, maybe the team is smaller overall, and then those who aren't on their way to being super senior move into more of a generalist role. I think that's like a reasonable soft landing, but I do think you're going to want to make sure you're investing not only in today's senior talent for each specific function, but in tomorrow's. Otherwise, I think you're going to regret it in a couple years. My take. That said, who knows what the world looks like in a couple years. So my big thing generally is don't be overly confident in whatever your predictions are, because there's just too much flux right now.
Benedict Evans was on the podcast recently, said the same thing. We don't know anything about what's going on.
Yeah. Yeah. I like him a lot. I'll make sure I listen to the podcast.
Yeah. So you talked about taste. This makes me think about, so you're interviewing a lot of people, hiring a lot of people. What are some traits that are trending up, things that you look for more and more now in this world? And what are some traits that are trending down and maybe less important to you?
I mean, there are some things that are the same, right? So for the longest time, almost no matter what the function, I always look for three things. Do you have sort of grit? Like you've got some drive, some fire in your belly. Are you a quick learner? And are you reasonably, ideally very, self-aware, so that you can actually take feedback and know what you're good at? Because if you're those three things, if you've got fire in your belly, you learn quickly, and you're self-aware, you can kind of get good at anything eventually. But if any of those things are missing, there's usually an issue. So that's sort of like the baseline right now for hiring. But just for people who I think are going to be more successful over these next five or 10 years as things change so significantly, I think two things that I'm continuing to encourage myself to do are to stay curious and to put yourself out there. I just think you've got to try things, right? This is, to that point before, that no one really knows what's going on. You just have to be willing to try things. It's almost... I don't know. Do you speak another language?
Russian. Yeah.
Yeah. So when you learn another language, I think one of the most important things, one of the best predictors, this is my guess, I don't have any research on this, about, you know, are you going to get good at
speaking is are you willing to sound like an idiot? Are you willing just to say it and be corrected and not be offended and then just get better and better? You just have to put yourself out there. And with all of these new tools and models and technologies, I think you just have to be willing to try stuff. So if you're curious and you try stuff, I think that'll, you know, you'll learn, you'll adapt. But if you're not curious or you're not willing to make mistakes or try things, I think you're in a ton of trouble. Or at least I think it's going to be a really difficult time.
So those I think are premiums, not just for hiring at a company like Meta or a team like Instagram, but I just think across the industry and multiple industries over the next 10 or 20 years.
Is there something that maybe we're looking for less of for some of these functions?
I think that there's some that are still going to be very large teams and so you need people who are really good at managing large organizations. Large organizational leadership is its own craft and skill. It's actually different than management. But I do think there'll be less of those roles. I think we'll have more smaller teams and there'll be less people who manage thousands of people. And so that's not that that job will go away, but that will be less of what I'm looking for in hires because I'm going to have less roles like that.
Something I'm hearing from a few folks is AI is almost kind of resetting people's impact and success in terms of some people that were maybe low performers pre-AI can now do like things they were bad at or AI now allows them to do and now they're thriving, building all these things, helping other people. Do you see that at all? Just like AI is just like lifting other people up, maybe lowering some people down.
Yeah. I mean the job is just different. I mean take engineering. Engineering used to be maybe not majority but a large percentage, 40, 50, 60%, writing code. You know it's not now, especially if you talk to anybody at these labs, they're spending most of their time planning and reviewing code. That is a very different job. You might hate that and you might have loved just writing code, or you might love that and you might not have been that fast at writing code. So you know who succeeds is a function of whose strengths are aligned with the tools' needs and the business's needs. And so this is definitely happening.
Another thing is you've had people who had good ideas about how to contribute to other functions but didn't have the mechanical or technical skills to do so, and AI reduces the boundary to do that and then all of a sudden they can't, like, you know. For me it's kind of funny because when I got hired at Facebook, all the designers had to be able to program. I went through a technical loop. We gave up on that because it was too hard to hire people. But I now get to program again for the first time in maybe 10 years and you know I am not a good engineer. I'm a mediocre engineer on a good day, but now I can write code responsibly, which is just an amazing thing.
You're seeing this across all sorts of levels and seniority and functions. You know, designers who are programming, engineers who are pulling data and doing strong analyses, data scientists who are putting together proposals for designs.
You know, the tools aren't all great, by the way. I think too often we have this really polarized binary outlook on the state of AI, like are you AI-pilled or are you anti-AI? It's like people aren't binary. I said that to the team yesterday. And the state of the tools isn't binary either. You know, they're amazing at some things and remarkably bad at others. And the people who I think are going to make the most of it are the ones who are clear-eyed about what AI is good at and what it's not good at, and also have an instinct or a nose for what it will be good at and not good at, you know, next month or in a couple months from now.
You mentioned that AI writes all our code now. Someone tweeted this idea that stuck with me for like months now of just like, remember we used to be able to just write code for free. [laughter]
I think you'll still be able to write code for free. It'll just be with a smaller model. But yes.
I guess that's true. Like there's models that are close to free, but it's like, yeah, that's crazy. Now it's just like
But just think about the cost. Think about what you pay for a model now and what the level of intelligence you're getting from that model is, and then at that same price point a year ago, what were you getting? At some point the incremental value won't matter. Like, you know, we're getting there I think with small projects and programming. I think the models will matter even beyond, you know, this week you've got Fable and obviously Mythos from Anthropic, but I spent a lot of time with that this week. It's for the first time I'm like, "Oh, I'm just talking to a much more technical, much smarter engineer than I am." You know, the next version, a year out of that model. Do I need to pay for frontier tokens for whatever Anthropic model 6.0 is, or is Fable just fine for all of my side projects? Probably just fine. Probably pretty cheap by then, too.
Yeah. When Kevin Weil was on the podcast when he was CPO at OpenAI, he famously said, "This is the worst the model will ever be."
Yeah.
Yeah. Still hard to comprehend that. Wow. It's only going to get better.
So, on this point of token spend, ROI, and things like that, Meta was famous for this leaderboard of token spend.
It's a terrible idea. No leaderboards for token spend.
Okay. Talk about that and just how do you think about budgets for engineers and product teams at this point? You just like spend as much as you want? Is there a cap? Is there any sort of thing you've kind of figured out that works well?
Right now we've managed to get the costs reined in a little bit by shutting down the silly things that we were doing. And so, you know, it's not that hard to build a token incinerator, and that doesn't create a lot of value. And as soon as you actually look at the dollars in and value out, you might just be like, "Oh, that's just a bad idea." And so right now we don't have token limits for our engineers, actually I think for anybody really. I think that'll eventually have to happen, particularly if costs go up before they go down. I think they'll eventually go down for the reasons that we just talked about.
But I think of it as any other resource, right? Like I have to decide how to deploy capacity to my different teams because I have a limited number of GPUs and CPUs and storage and RAM, etc. I have to decide how to deploy opex for labeling budgets across my teams. I have to decide how to deploy payroll for headcount across my teams. I think that you can imagine, at least in a year or two coming, that the burn rate of a strong engineer might be the same as their salary or their cost of employment. And in that world you're going to probably need to put in some caps. The cap should probably be proportional to the company's trust in your ability to use them in an ROI-positive way, but I can imagine caps being healthy. Right now we're not there. I think costs will go up because we'll just be using more tokens, not because prices will necessarily go up, but then I think prices will come down because all of these frontier models are going to be in a bit of a pricing war. So we'll see. I think it'll be a bit of a roller coaster.
So, coming back to this idea that, as you said, we've evolved from we used to write all our code to now we're approaching all code will be written by AI. And it feels like now the transition is it's not just written by AI, but it's like one-shotted by AI. Like coding now is steering AI, and how often you have to correct it is coding now. And so it's like the software development life cycle slowly being eaten by AI. It'll start helping us come up with ideas, I imagine, more and more. The question I like to ask people is where do you think human brains will continue to be most valuable as AI continues to eat more and more of that product development life cycle?
Taste, like we talked about. Judgment, particularly around strategy, right? You might get feedback from an AI on a strategy, but you're not asking an AI to come up with a strategy anytime soon, or if you are, then it's within the context of bounds you set. So here's my goal, here's my vision, here are my constraints, here's my job, here's my budget. I think that it looks more like management, right? Like you are trying to define what success looks like, decide how prescriptive you want to be about the path to success, and then giving feedback along the way. And that is its own craft.
And it'll be interesting to see how some of the same dynamics come up. Like I believe that if you are too prescriptive as a leader with a team, you end up stifling good ideas, but if you're too open-ended, sometimes teams just waste time going in the wrong direction. And so that level of autonomy you give a team, maybe that applies to agents in the future, particularly when we're talking not just about building something, but deciding what you build in the first place.
But I think of vision as an articulation of the world or the state of the product you want to get to. Then I think of strategy as an opinionated path to achieve that vision. A strategy can't be like "be the best" or "be amazing." It has to be controversial, that a reasonable person should be able to disagree with it, because otherwise you're probably just trying to compete on raw execution. And I think both vision and strategy are going to be where our brains are spending more and more of our cycles, and I think less on execution.
Something I have always thought is AI should be incredibly good at strategy. Because you would think, here's the market, here's all the information on the market, our competitors, our metrics, our numbers, our growth, all these things. Help me figure out how to win. You'd think AI knowing all that would be really good at this.
I think it could be. I have found it's not unless you steer it pretty aggressively. And I don't mean towards an answer. I mean based on the constraints. It turns out when you're trying to come up with a strategy, there's a lot of things to consider, right? You need to consider the state of the technology, the personnel on the team and what's motivating them and what you can get. You know, sometimes coming up with an idea that is on the bubble is going to actually attract some of the best talent, and so that kind of push then goes to the idea. Obviously the competitive landscape, the regulatory landscape for companies as large as ours, and the compliance landscape, the identity and reason to exist for the brand. You have to consider all of these things.
I think if you ask an AI just for a strategy lazily, you're not going to get something great. You're going to get something pretty predictable that probably the competition would expect you to do. I think if you want a more effective one, you need to think long and hard about what are all the different inputs that need to be considered. Make sure you steer the AI in a way that it's considering those as well. And it needs to be a conversation and a back and forth. But I think if you're willing to put in the work and the time, it can definitely be helpful and definitely be clarifying, particularly if you tell it to be critical. Different models have very different vibes though on how willing they are to push back. So [laughter] I recommend picking one that likes pushing back.
Yeah. Mythos has gotten really good at being like, I can't do this. Let's move on. Like there's all these
fun has always been a little bit of a jerk in a way that I actually appreciate.
I truly
I appreciate it. I really do because I don't want one that's just like, "Oh, you're so right. I'm so sorry I said that." It's like, "No, hold on." I want the real sort of intelligence. I don't want the pleaser.
This point you made about people being excited about the strategy is such an interesting one. There's this idea that I read, I think Cory Doctorow wrote this. There's this concept of a centaur and a reverse centaur. So a centaur is a human body, horse. This is going somewhere, I promise. Human upper part, lower part
Horse body. Yeah. Yeah. Yeah.
Horse body, where the human is in charge. And we prefer that, we want to be in charge. Reverse centaur, which is what we want to avoid with AI, is where the AI is controlling us and we're just doing its bidding.
It's a horse head on a human body.
Yeah, exactly.
It's terrifying.
Like in a sense, Uber drivers and DoorDash people, this is kind of their life, which is not great. [clears throat] And this is the danger for a lot of people. If it's giving us the strategy and telling us here's what we're building, no one's going to want to do that. So that's a really interesting counterpoint, that we don't want AI to be telling us the strategy, almost.
Yeah. No, I think there's a lot of things to be careful about right now. And I would certainly not just assume that because you might be able to outsource some workflow to AI that you should. There are certain ones where I think it's really just a win-win. There are certain ones where I think the risk outweighs the benefits.
This episode is brought to you by Mercury. Radically different banking loved by over 300,000 entrepreneurs, and now with Command. I've been a customer of Mercury's for over 6 years. I have never once thought about leaving. Mercury is basically what happens when banking is built by product people, not by bankers. [music] They make it so easy, dare I say fun, to send invoices, move money around, set up virtual cards for folks on my team. Does your bank have an API, a terminal [music] native CLI, or an AI-ready MCP server? I don't think so.
And just recently, they launched Command, a conversational interface built directly into Mercury, which acts as your financial operator. I've been using [music] Command to transfer money around, to figure out what categories I've been spending the most money [music] in, analyze my cash flows, and just today, I used it to find out how much I've made from a specific sponsor over the past year. I just asked, "How much have I made from X over the past year?" 10 seconds later, I have an answer. It [music] is so freaking cool. Visit mercury.com to learn more and apply online in minutes. Mercury is a fintech company, not an FDIC-insured bank. Banking services provided through Choice Financial Group and Column N.A., members FDIC.
Okay, going back to product leadership, things you've learned along your journey. We were chatting ahead of this about just things you've learned, and one thing that you said about some of the best product leaders you've worked with is that they're less visionary and more curators. I'd love to hear more along these lines.
Yeah, I mean you do sometimes find amazing product leaders who are just like idea machines, just prolific idea machines. But I do think a lot of the best have taste, have something about them that really makes strong talent want to work with them, but end up sort of being curators. Curators of people, curators of ideas, curators of technologies, curators of strategies, because I don't really care, if I'm hiring a strong lead for an area, if the strategy comes from them or comes from somebody else. I just care that there is an amazing strategy and everyone has bought into it and that the and then
we're executing against that strategy well. And so I think that some of the best product leaders, yes, have ideas. It's hard to be a great curator if you don't have some of your own ideas, but are that embrace the reality that they can't come up with everything themselves. And so they need to create an environment in which great ideas bubble up and are chosen or decided upon. And so, you know, I think it's not just about curating ideas, but it's sometimes about curating teams and people.
I love that. I so agree. I feel like everyone's always joining a team and they just want to do vision strategy, just like not actually hands-on work, and now AI is coming in here: let me do the strategy.
Exactly.
And I love this point that there's so much power and value, and people underestimate just the need for just like a really good curator of the team's ideas.
Yeah. Sometimes it's also not just who's good or what idea is good. It's also what is going to work given the broader context. So for instance, on team building, a huge thing that I'm always considering is not just, is this person a really strong candidate for this role? It is, how does this person fit into their leadership team?
You know, so for an area like trust and safety, you know, I have an engineering lead, I have a product staff lead, I have a data science lead, I have a design lead, I have a research lead. You know, I need to make sure that those five complement each other, and that's, you know, about what skills each one has, what weaknesses each one might have. I also need to make sure that they, and this is more art than science, have a good vibe, right? You know, you need trust and rapport. A leadership team with strong trust and rapport can work through most anything. A leadership team without trust or rapport, like anything can become an issue. And so that chemistry bit is, like I said, much more art than science, but that also matters. And so I think some of the best leaders, and product leaders specifically, also either do that instinctively or consciously, but, you know, they have a nose for building teams that are going to have good energy and good collaboration.
Warm and fuzzy stuff.
Yeah. Yeah. Well, the flip is also true, right? Like, I've had many times in my career two people who I think are amazing, and I even adore them and love them, and they just can't get along. [laughter] You're like, this isn't a competency issue. This is just a personality issue, and you just have to sometimes call it and split them.
I want to transition to talk about Instagram, the product, the platform, things you guys have learned there. Let me start with this question. What's something that the Instagram algorithm knows about human behavior that people may not realize?
One of the most common misconceptions is actually in the opposite direction. I think people assume that there's a much more detailed semantic understanding of everybody's interests and preferences in the algorithm than there is. Most of what's really driven the progress in the world of recommenders over the last 5, 10 years have been, you know, these large embedding models and these other techniques that basically produce artifacts that cannot be read by people. They're not legible. They're like giant vectors. It's like, sure, I can show you the vector, but it's just going to be a bunch of numbers in like a seven-dimensional space. And so when we talk about, does the algorithm know something, usually we think in these more semantic terms. It knows I like surfing, and it's like, it doesn't. It just has this big-ass number that happens to correlate with surfing.
That said, I think that is starting to change, right? I think that one of the things that LLMs are enabling is they can describe in, you know, words, English or whatever language you prefer, what some of those previously illegible artifacts are at least approximate to, if not mean directly, right? So this is like the thing I've been really—I posted about this this week—this thing called Your Algorithm. Basically the idea is we take a look at all of the stuff that you've interacted with, and then, you know, all of that is in an embedding space. You can think of embedding space as a map. You can map a bunch of videos into the same map, and so videos that are close are similar. And now we can just have an LLM just be like, describe that part of the map. And it can be like, oh, that is like deep pour-over coffee snobbery. And that's kind of amazing.
That is so cool that like you can ask the LLM to look at these numbers and extrapolate: here's like the topic that you're interested in.
Yeah. Or look at the videos. And so the way you both—so you can also embed concepts into that same space. And so, I mean, embeddings are really the underlying technology underneath LLMs, right? That's how the whole thing works. And so, you know, what we do now is we let you, you know, quote unquote, see your algorithm. You can see what topics we think you're interested in, and you can adjust it. You can add and remove things. But the idea here is giving people some agency back in a world where, you know, these social media apps are getting taken over by recommendations.
But we can't do a lot of other things yet, which we will be able to do. You know, there's things that aren't topical that you might ask for. I want more fun content. I want to see my friends more. I don't want to see my high school friends' kids' photos. You know, I don't know, we can come up with, I don't want to see seven photos in a row, but I'm happy to see six photos in a—whatever your mind can come up with. So, we have a lot of work to do, and so I'm excited about that. But I think a misconception historically is, until recently, we don't really know as much about you as you think. We were just like, oh, like you liked these photos, these people also like those same photos and they like these other photos, so you might like those other photos. Like, I'm oversimplifying, but that's kind of how it worked. Only now are we actually getting as sophisticated as I think people have assumed we've been for many years.
That is really interesting. One that comes to mind is kind of this transition everyone eventually goes through to this like algorithmic, broad, global feed. It always feels like people think, I just want to see chronologically everyone I know and follow, and that's going to be my favorite feed, and it continues to be proven wrong. No, you actually engage a lot more when it's this algorithmic feed of things we think you will love.
Yeah, it's tough because, I mean, I posted this week this thing about agency and I just got destroyed in the comments, which is just part of the job. I get it, right? But there are a couple issues with the chronological feed. So one is, and some of this is the tension between an individual's interests and what works when you scale it up, right? So if you do a pure chronological feed, the incentive for everybody is to just post as much as possible, because it will always be at the top of everyone who follows you's feed as soon as you post. So what ends up happening is that the feed gets overwhelmed with professional content, with usually large company content and publishers, because, you know, the New York Times can pump out 50 things a day. Your best friend won't. You might get one thing a week from them, and so your feed just gets taken over.
So part of it is the incentives that emerge, because when you design these systems, it's almost like designing a city. You need to think about, okay, here's how the mechanics work. What are the incentives that arise? How are people going to act within those incentives? And then what happens? And the other thing is, sometimes the most interesting thing is just not the most recent thing. Recency is an important input into relevance, but it's not the only one. My sister got engaged last night, and you know, she's in Germany.
No, she didn't. If she did, she's married.
She got married last year. That's why it was top of mind.
All right.
But if she got engaged, you know, and I missed it because she lives in Europe and, you know, we're in different time zones. Like, do I really want to see a picture of my brother's po' boy sandwich, or do I want to see my sister's things first? So it's tough. It's tough.
I'd love to figure out a way to find the right balance. I want to give people agency over the experience, but I think it needs to be in a way that creates a system that makes sense, not just for us as a business, which matters—I'm not pretending that's not an issue—but also for the overall community. Because we've done chronological by default, and where you can make it default, and you see not only does usage go down, overall sentiment goes down. The individual who made that choice might be happy at the moment, but when you just get pummeled with stuff you're less interested in over the course of months—we ask, we run surveys at massive scales—we just see people start to become less and less satisfied with Instagram.
Kind of along these lines, everybody asks you about this these days, AI and content and how that all impacts everything that's going on. I want to ask you something I haven't seen someone ask you. Is the rise of AI content a headwind or a tailwind for Instagram versus other platforms? Do you think this helps or hurts you guys?
I think it's going to be a tailwind, but I think it's going to be a challenge. And not just because it's more content. Obviously, we're an attention-driven business. We're an advertising business. More content means potentially more attention. That's not for free, though. Like, I don't think we're very good at ranking AI content yet. There's great AI content. There's crap AI content. You should just see the stuff you're interested in and not any of the stuff you're not interested in. But I do think that in a world where, for years now, and I've said this many times, power is shifting from institutions to individuals across industries—the easiest example is sports, where players are more relevant than teams now, and that was not the case when I was a kid.
In that world, I think it behooves us to invest in individuals, and to invest, specifically for Instagram, in creators. And I mean creators broadly. I don't just mean influencers who are promoting branded content and making, you know, native-only videos. I mean anybody who's using platforms like Instagram to help do what they do, right? You could be a journalist, you could be an artist, you could be selling scarves you sew, but like you're out there as yourself, creating and sharing content that helps you achieve whatever it is you're trying to do. So we've been leaning in that direction for many years now. That's been one of our two or three most important audiences for as long as I've been on Instagram.
In a world where there's an abundance of synthetic content, I actually think people are going to seek out creativity and authenticity and people more, not less. And I think that that will help us. That doesn't mean that we won't have AI content on our platform. There's going to be bad and good AI content, and we're going to try and handle that, you know, the way we normally handle content. So unsafe goes away. Interesting versus not interesting is based on ranking and personalization. But I think people are going to really seek out other points of view, because Instagram was never just about the content. It was always, to a certain degree, about the person behind the content, the point of view, the reason they're sharing it, their perspective. And I think that's going to become more important, not less. And I think given that we are not the best at a lot of things, but we are the largest creator platform, you know, if you look at how we define creators and how many creators use us versus other platforms, I think it'll be a tailwind for us, because I think people are going to seek out people.
And this connects to your earlier point that companies like, say, the New York Times can pump out a bunch of AI content versus a creator. And so you're saying you kind of want to protect against that too, allow individuals to continue to perform well in spite of all this AI content.
If you just love AI content, great. Like, you should be able to have a feed that's just like AI town. And if you don't, then you shouldn't have it in your feed. Like, you know, to me, I understand why people are, right? I'm not oblivious to the overall paradigm shift and sort of revolution that we're sitting in. But I don't think we should judge content based on the tool that made it. I think we should judge it based on the content, the point of view, the person behind the content. I don't think we should filter out AI content. I think we should let you know if content is AI content or not. I think we should let you know more about the person who posted anything, so that you can make informed decisions about whether or not to believe or trust them based on, you know, knowing who they are or where they are or how many times they've changed their profile, or if their profile is three days old or three years old. But I don't think we should be making value judgments based on what tool you used.
Is there an AI content creator you love that you're just like, "This is so good. I love watching these AI videos."
Yeah. What is she called? Plastic Dream Sequence. Is that what it is?
I think—I'll check the name. Yeah. Plastic Dream.
Plastic Dream Sequence. I have it on my phone. I'll double check. It's these like sort of dolls, Barbies, but they're like singing songs, and these little tiny silhouettes and snippets. And it's just amazing. It's a little weird, but also kind of amazing. And it's very clearly AI. It's not pretending not to be, but it has a very clear creative and aesthetic point of view. And every time I come by one, I'm like, "Yep, we're doing this now. I'm going to watch this for 30 seconds."
I have it pulled up here. And I want to watch it, but I'm not going to. [laughter]
That's awesome.
If Only AI, that's another one. He's in France. I think he's in Paris. He uses multiple different tools and models, but he kind of tries to create these dreamscapes and animate them. So he uses one model to create the image, another one to create the video, music, etc. He's very clearly got his own aesthetic, and you could think of him as a painter, but like this is his tool.
Is there kind of a vision of AI versus human in the feed? Like you said, you maybe want to mark—how do you think about people? Are they going to be like AI account, not AI account? How do you think about it, or is that still kind of a work in progress?
Maybe we'll end up in the same place, but there's a difference between marking content and marking accounts, and they're both useful and interesting. So if content was created with AI, I think you should be able to know that. That's hard, by the way, because we can detect that right now, but as these models get better, we might lose the ability to detect that. So we should also be very careful to be honest with you about how confident we are in our own sort of assessment. But I think you should be able to just ask, "Hey, is this AI?" And we should be able to tell you: we think it probably is, or we're not sure, or it's definitely not, or it definitely is. I actually think it might be more practical to label camera-captured content, basically non-AI content, as opposed to labeling AI content long term, for a couple of reasons.
But then at the account level, I think
it also matters. There is definitely a new spam vector, which is these fake accounts. By the way, an AI creator, that's fine, there's nothing wrong with that necessarily, but there are these spam vectors which are trying to abuse that, and, you know, they're selling bogus supplements, and it's like an AI monk, and it's not obvious that it's an AI, and it's just trying to take advantage of a certain aesthetic or a certain sort of stereotype. We need to figure out how to crack down on that. And so I do think we should be making sure that, basically, you just need to know, and then you can make your own informed decision. Is the account a real person or not? Is the content a real piece of content or not?
When you think about other platforms in the space, social content platforms, are there any features or ways of approaching stuff that they do well that you're kind of jealous of or really impressed by?
Yeah, there's a bunch. Everybody. So many people do so much. For me, one of the things that we are finally catching up with, but I've always been very impressed with, is TikTok and their recommender's ability to break small talent. In the world of ranking and recommenders, you can talk about exploitation-based ranking. That sounds terrible, but it just means using the data you have. And then you can talk about exploration-based ranking, going and trying to figure out what someone might be interested in that they might not even know they're interested in yet. And it is much easier to move engagement by showing people stuff that they'll probably like because lots of people like it. It's much harder to go and figure out how to essentially test content so that we can see, like, hey, sure, you like Bieber, but you might also like Afropunk, and so we're just going to show you some Afropunk and see what happens.
If you do the latter, this exploration-based ranking, I think it's really good for niche creators and small creators because you give them a chance to find an audience that either wasn't going to see them before or didn't even know that they were interested before. So we've invested a lot over the last couple years in ranking, not just increasing engagement, but increasing originality, increasing the number of pieces of content that break out, increasing recency to stay culturally relevant. And so a lot of that has been inspired by TikTok and ByteDance. I think we're catching up. There's actually a couple of those areas where, best we can tell, we're ahead of them. There's a couple where we're still behind, but we have line of sight to, I think, being best in class at recommendations for the first time during my tenure. So they get a lot of credit for inspiring a lot of their work.
Nice job.
Well, we'll see. We're not there yet. They call me disappointed dad. My team is always like, "Can you ease up on the disappointed dad vibe?" So I'm trying to be a little bit more generous about giving people their flowers.
Like, you say that, but that's an interesting common thread across really successful leaders, just never being satisfied.
Yeah.
Always feeling... a blessing and a curse.
Here's all the problem. It is. Yeah.
On this creator piece, I think that's also, you know, people complain about this global algorithm not showing them all their friends, but I feel like this is a benefit of what happens when you do this now, that you can break new creators into a wide audience if you have this kind of global algorithmic feed, which is really great for a lot of people.
I mean, I'm out there talking about a lot of these contentious issues and I get beat up a lot in the comments, which is fine. My main thing here is just to try to communicate that there's almost always trade-offs, right? You can't just have all of the things, unfortunately. You want to never see something you're not interested in, then you're also just going to see the most basic, general, lowest common denominator stuff all the time. You want to discover new and interesting things, you're occasionally going to see stuff that was just a miss. But this isn't just true about ranking. All these major debates have trade-offs, right? Privacy and safety, those two things are in tension. Do you want a company scanning your messages or not? There's some really significant trade-offs on both sides of that debate. And so generally speaking, when I argue and engage in debate with people who feel really strongly about things, I'm not usually trying to convince them. Their mind is usually made up. I'm just trying to enumerate all of the different puts and takes for the rest of the people watching the conversation.
Speaking of getting torn apart in the comments, you're so in the middle and thick of all of these really hairy situations, changing the feed, you're in the Cambridge Analytica lawsuit, all this. You're in the center of so much controversy, and is that some—
Yep.
Oh man. Is that just like you, "I will lean into this, this is the thing I need to do"? Or was it like Zuck being like, "Adam, you got to be the front face of all the stuff and get in there"? Like, where does that come from?
It started on News Feed. So I used to run News Feed at Facebook, and my take was that the debate was going to happen with or without us, so we might as well participate. And so I started being really active on Twitter specifically because that's where journalists really lived at the time, and I thought it would show some humility to show up on their turf, so to speak. My Twitter ended up being the darkest place in my life because I just followed all of our biggest critics. That's not a dig on Twitter. That was just what I did.
And that's where it started, and it kind of slowly built from there. For better, for worse, we've become a really important part of daily life for a lot of people. We touch a lot of people. We have a lot of responsibility, and there's a lot of change, and change means there's going to be anxiety and stress and scrutiny. We've made great decisions. We've made mistakes. We've been criticized unfairly. We've been criticized fairly. And so we just need to accept that this debate is gonna happen broadly. So I just think it's better for us to talk about it and just be clear about what we're doing, why we're doing it, what the trade-offs are. If people disagree, that's okay. We're not necessarily winning over friends when we talk about what we do. But I think over the long run, people are fundamentally more afraid of things that they don't understand and about things where people are more secretive and less accessible. And so I've tried to show up in an accessible and authentic way. And I've made mistakes, and I have enjoyed it at times and hated it at other times.
But that's kind of how it started. There was also kind of a fun debate in Mark's senior leadership team a long time ago where we were just talking about how we're a social media company where we had a very conventional approach to communication, like press releases, and why don't we just use our platform. I was not in that debate, but I stuck myself into that debate, trying to mediate it. But that was also a reason why I ended up getting sucked in, because Mark was like, "All right, well, why don't you try and see how it goes?"
What's something that helps you deal with the hate that flows at you every time you say something that people disagree with?
You try to put it in perspective, right? So it started with, I did the redesign of News Feed in 2009. We launched it March of 2009 for Facebook. I was a designer, like an IC designer, entry-level designer. And the first comment that came in was something pretty derogatory. I think it was homophobic and anti-Semitic. Literally, we're all sitting there, we launched this thing and we're just looking at the stream of comments, and it's the first one, and it was specifically about... they don't know me, but it was like, "What expletive, censor, censor, designed this?" And I was like, oh, it was me.
And I was devastated. I was a 25-year-old kid, and I thought about it, and I came to this idea that if you spent 30, 40, 50 minutes a day at your desk, and you organized your photos there and you wrote letters to your friends there and you read there, and then I just came and I rearranged your desk, and I didn't tell you, I didn't warn you, I didn't even explain why, you would be pissed, and that would be reasonable. And that was what was happening, just with millions of people. So I try to put things in perspective. And then I try to step away from it, get time with my kids, get time outside. There are months where it's really not hard at all, and there are months where it really grinds on me.
Along those lines, there's a famous kind of reversal when you redesigned the feed into this kind of video scrolly experience. There's this whole—
Protest. The world protested.
Yeah, that was pretty rough.
What was kind of like, okay, wow, we're actually not right and we should go back? What helped you decide, okay, let's change course?
So actually, three or four things got conflated on that one. We had a redesign of feed that went to the video viewer that was a test to 4% of users on iOS. It was not going to roll out. It was just an early test to get some sense and feedback on the idea. We were also leaning into Reels a lot. We were also leaning into recommendations, so posts from accounts you don't follow, a lot. And there were also creators who were upset about the fact that their reach was going down and they were blaming ranking changes for that. Those four things got all conflated. We had some pretty big-name creators publicly slap us. Then the press covered that creator backlash, which then got more creators doing it. So we ended up with a little bit of a multiplier effect or echo between the creator community and the press back and forth, but we were never going to launch that. That was an early test. We knew it needed work. We actually have continued to grow video and invest in creative tools and invest in ranking and invest in recommendations, and that's driven most of our growth in the years since.
I think we were pushed... well, I don't know, doesn't feel like it, but I think my real takeaway wasn't that we should not have tested that design necessarily. I think we could have done a bunch of things better to explain, and maybe move a little slower. I think we were just pushing things a little bit too fast, and when you are responsible for a platform like Instagram, you need to be reasonable and realistic about how much you can evolve it. Now, I would much rather have backlashes like that every couple years and continue to evolve and continue to stay relevant than the alternative, which would have been we didn't have video, we didn't have DMs, we didn't have Stories, we didn't have ranking, and we wouldn't be having this podcast right now. But the cost of leaning in is that you're going to occasionally make a mistake, and you're going to definitely pay for it.
It's interesting how running experiments now is very risky for companies at your scale. One person spots it and they're like, "Oh shit."
You kind of need to have a press... You don't need to be proactive about communicating it, but you need to have a comms strategy. For any design change or any test that could be controversial, we talk about it beforehand and are like, "Okay, not if it leaks, when it leaks, what are we saying? Should we talk about it proactively, should we talk reactively? Either way, what's the message?" Because you can't launch something to three billion people and not test it first, but you can't test something at our scale and not expect people to cover it. So you have to be ready to talk about it before you even know you want to launch it. So it makes the development cycle more complicated than it used to be.
Yeah, the head of growth at Anthropic launched an experiment with pricing and it just went crazy on Twitter. He's like, hold that, 1% of people were just trend stuff like noic.
Pricing particularly, that one is a real... you got to be real careful with that one. I've learned, we've all learned these lessons. We should all share notes more. That's what I think.
Here's one note: how to avoid the internet hating you for the day.
Yeah. Yeah. I'm happy to talk to that more than...
I think he's all right. He's all right. Okay. Okay, I'm going to take us to two recurring corners on the podcast: fail corner and hot seat corner. Fail corner. What's something that you worked on that was just a huge failure that helped you become better?
Oh, a bunch.
So any—
So I'll give you two, maybe. So before Instagram, my first project as a PM was on a project called Facebook Home, which was a sort of fork of Android at the operating system level and a piece of hardware with HTC. It was a spectacular failure. I learned way more in that year, year and a half, than I did in any year, I think, probably in my career, because I was just a design manager before that. I declared myself a PM because the PM on the project quit, and I just threw myself head first into understanding carriers and OEMs and certification as well as Android and operating systems, and just learned a ton. And I'm happy I brought that project to an end because it had been going on for a long time, and sometimes the best thing you can do is execute an idea that doesn't have market fit well, just to decide whether or not the idea was a good idea in the first place.
Another big mistake I made during my Instagram tenure was the first version of Reels was built on top of Stories. Stories had a ton of momentum. This was, I think, 2019, and we were trying to build Reels into Stories because we were trying to build into the thing that was growing the fastest. But it was not a strong foundation. The read-through rate on Stories is relatively low. There's way more stories than most people have time to consume. So most of the reels were never seen, and then they disappeared. And if we had the version of Reels that we launched in, I think, the summer of 2020, in the summer of 2019, I think TikTok is still big and important, but I don't think it's as big as it is now, because when they really took off was when the pandemic hit and a bunch of people had a lot of time at home and were looking for a little bit of joy and were totally fine with their phone having sound on. And so if you look at the numbers, 2020 is when they exploded, and we were out of position. And on one hand, I'm a designer. I'm trying to not add new things to the product. I'm trying to extend existing primitives, and that was the idea. On the other hand, I was wrong, and it's a pretty big fork in the road if you just look at the overall business over the last eight years.
We created a lot of economic opportunity in the world allowing TikTok to grow.
That's good. I'm glad they exist.
Okay, final question. I'm curious just about your screen time policy with your kids. I know you have three kids. There's a lot of concern these days about Instagram not being great for people, not for kids. A lot of tech
executives don't let their kids use devices while they're building the product. As head of Instagram, how do you think about screen time with your kids?
The key thing for me is boundaries. It's also about education and being and having conversations with them. But my kids are too young to use social media. They're 10, eight, and six. But they each have an iPad. They get, they have to earn their time. So they have different ways they earn that time. It's usually about like sitting down to do your homework three times for half an hour each gets you a total of 90 minutes on the weekend and then they can use that time on the weekend. But you kind of have to set that boundary where it's like you can't just, you know, it can't just be they ask for it and you give it to them. I think that matters a lot, and then I'm pretty opinionated about what they do on it. Like I approve what apps they have. I think parents should be approving what apps kids specifically are downloading onto their devices. We've been advocating for this at a policy level for a long time at Meta. I think those things help a lot.
There are some exceptions. One is planes. It's just about surviving. I don't know if you've ever, for those of you who are parents...
I'm going to be. Yeah.
Yeah. It's like you just, that's just like, all right, you know, it's a 10-hour flight or eight-hour flight. It's like, yeah, you just need to get through it.
The other one that I'm starting to experiment with my 10-year-old with is, so schools are interesting because I think I'm pretty supportive of no phones in classrooms. That's happening more and more. I think that is probably good for education. And I do also know in the world of AI that there's concern about kids using AI and not learning critical thinking skills. And I think that's a valid concern. But I also am worried about kids not learning how to leverage AI and then being sort of at a disadvantage. So that's a balance. So I think you need both.
So with my eldest, we started vibe coding recently together. He just loves video games. So I was like, "All right, let's make a video game." And so he's made this 19-level platformer game that kind of looks like an eight-bit version of Super Mario from when I was a kid. But each level has its own theme, its own types of monsters. There's a store where you can buy different skins or weapons, and it's unbelievable what a 10-year-old who still types with three fingers can do with just, you know, a couple hours of sitting together. But that is more of like a, I want you to learn how to make things. I want you to be thinking, not just playing games, and I'm going to sit with you and we're going to do this together.
So to me, these are the things that matter. Boundaries, scoping it down to the activities you think are healthy for your kid. Every kid is different. But I do think you want your kids to be digitally literate, AI literate, because I think if they're not, they're going to be at a disadvantage. But you also don't want it to be a free-for-all.
This is selfishly useful for me as a, I have a three-year-old and I'm trying to figure all this stuff out. So, this is useful.
Oh, yeah. No,
we need to figure out a strategy.
It's a thing. And you're not that far off. You're really just not that far off. It's going to happen in a couple years.
Vibe coding next year. Let's do it.
I couldn't believe it. I tried to do it six months ago and it just totally didn't work. And then now with newer models, it's been amazing.
What's their platform of choice? Are they a Claude Code person?
Yeah, my 10-year-old is using Claude Code right now. But we will see. We'll see how that goes.
Adam, I'm going to let you go. Thank you so much for being here. You're just such a gem of a person. It's just so obvious how authentic you are and just how deeply you think about everything. So I really appreciate you being here.
I appreciate you bringing me on. Been a fan for a long time. It's nice to finally get to have a conversation.
I appreciate that. I really appreciate that. Let me just ask you this final question I ask everyone. What's a way that listeners can be useful to you?
I just think, you don't even have to tell this to other people, but just remember that this world and technology is complicated and there are almost always tradeoffs. And you can totally disagree with the decisions I or we make. But just remember that we are people here trying to make these decisions, trying to do the best we can. And I actually do invite the criticism and the critique and the feedback, but know that none of these contentious debates are nearly as simple as most people pretend to make them out to be.
Wise words. Adam, thank you so much for being here.
Pleasure. Thank you, Lenny.
Bye, everyone. Thank you so much for listening. If you found this valuable, you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app. Also, please consider giving us a rating or leaving a review, as that really helps other listeners find the podcast. You can find all past episodes or learn more about the show at lennyspodcast.com. See you in the next episode.
Article published · Updated