Adam Mosseri: AI is a tailwind for authenticity
In a Nutshell
Adam Mosseri argues that AI boosts productivity but makes human taste, judgment, and curation more valuable than ever. Instagram's product teams are shifting to small "pods" with generalist product staff roles, while the platform will label AI content rather than ban it because authenticity will become a bigger differentiator. Success with AI requires being clear-eyed about its strengths and limits, staying curious, and focusing leaders on vision and strategy over execution.
These notes were generated by AI and may contain inaccuracies.
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 will make the most of AI 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 in the future.
People assume that there's a much more detailed semantic understanding of everybody's interests and preferences in the algorithm than there is.
AI content is going to be a tailwind for Instagram, but it will be a challenge in a world where there's an abundance of synthetic content. People are going to seek out creativity and authenticity and people. Instagram should not filter out AI content, but should let users know if content is AI-generated.
Adam Mosseri is head of Instagram. Over three billion people use Instagram monthly, which is one in every three people alive. Prior to Instagram, Mosseri designed and led the early Facebook news feed. He also ran the team that built the Facebook ranking algorithm. Eight 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.
The canonical team at a big company like Meta has changed significantly. For the longest time, the structure was 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—on the order of a baker's dozen. This structure existed because you want someone who can review code who is familiar with that codebase, and having these different specialized functions.
This year, Meta has adopted what they call pods, which are mini teams of four to six engineers who are more generalists. There is one product staff role, which is an evolution of the PM—a PM who can do some of what a designer does, some of what a data scientist does, and some of what a researcher does, leveraging the latest tools. Then they add whatever specialist they need: a senior data scientist for pricing strategy, or a very senior product designer for novel experiences. The team is built based on the needs of the work, resulting in a much smaller core of six or seven people.
By virtue of having less people to coordinate, pods can often move faster and make better decisions with less design by committee. While AI is adjusting and improving productivity, small teams are often just more effective.
On a team of six to seven, you have less specialists. You might have four engineers and a product staff with no data scientist, no designer, no researcher, and no content designer. The product staff is the generalist that supports all of those things. All functions are starting to bleed into each other, and the industry is wrestling with what that means.
A lot of what a data scientist does at a big company is relatively mechanical. Things like basic waterfall analysis—looking at all the steps of people creating reels and how people fall off on each step—are now much easier to pull automatically with internal tools, whereas a year ago this required a data scientist doing bespoke work. A product staff might be able to do that now.
Taste matters a lot. In a world where it's easier to build things, it's more important to make sure your time is spent figuring out what you should be building in the first place. A lot of designers are anxious about their roles because other generalists, engineers, and product staff are doing design. However, designers tend to have taste, which is something much more difficult to imagine being automated away.
AI-generated design work often has recognizable vibes—you can tell when something is a Codex app, a Claude app, a Replit app, or a Lovable app. Design should be thriving, but job numbers for designers are kind of flatlining. The missing piece may be the PM piece of deeply understanding the business and what will grow it.
Some designers will transfer into product staff roles. If you're an amazing designer, you probably have strong opinions outside of just interaction and visual design—you likely have strong opinions on product strategy, the business, and go-to-market. Some of the strongest product staff will be converts from design and from data science who are looking to expand their reach. They were influential across functional boundaries before, but blurred functional boundaries now allow them to jump in.
Strong designers will likely be technically generalists on paper, but they'll clearly have a uniquely strong ability in one type of craft with the ability and strong opinions to make informed decisions across other parts. Not all the strongest designers will stay in design roles, though the majority probably will.
This is probably the age of the generalist, but there will still be a real important role for amazing specialists. There is a GM model where different types of functions can become GMs, and the product staff role feels like a similar situation. This was true of PM before, but it's just so much more true now.
There's anxiety everywhere about competition, job displacement, and unintended consequences of technology moving so quickly. Functional lines will continue to blur, but there will still 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. You need to hire and mentor and grow talent. If you're not investing in tomorrow's senior talent for each specific function, you'll regret it in a couple years.
For the longest time, almost no matter what the function, three baseline traits are looked for: grit (drive, fire in your belly), being a quick learner, and being reasonably or very self-aware so you can take feedback and know what you're good at. If you have those three things, you can get good at anything eventually.
For people who will be more successful over the next five or ten years, two additional traits are staying curious and putting yourself out there. You have to be willing to try things. When learning another language, one of the most important predictors of getting good at speaking is whether you're willing to sound like an idiot, say it, be corrected, not be offended, and just get better. With all of these new tools, models, and technologies, you have to be willing to try stuff.
There will still be some very large teams, so people who are really good at managing large organizations will be needed. Large organizational leadership is its own craft and skill, different than management. However, there will be less of those roles because there will be more smaller teams and fewer people who manage thousands of people.
AI is resetting people's impact and success. Some people who were low performers pre-AI can now do things they were bad at and are now thriving, building things, and helping other people. The job is just different. Engineering used to be 40-50-60% writing code; now it's mostly planning and reviewing code, especially at AI labs. Who succeeds is a function of whose strengths are aligned with the tool's needs and the business's needs.
People who had good ideas about how to contribute to other functions but didn't have the mechanical or technical skills to do so can now contribute because AI reduces the boundary. Designers are now programming, engineers are pulling data and doing strong analyses, and data scientists are putting together proposals for designs.
People aren't binary about AI—you're not either AI-pilled or anti-AI. The state of the tools isn't binary either. They're amazing at some things and remarkably bad at others. The people who will make the most of AI 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—not just next month, but in a couple months from now.
Meta shut down the silly things they were doing that were token incinerators without creating value. As soon as you look at the dollars in and value out, some ideas are just bad. There are no token limits for engineers or anybody right now. That will eventually have to happen, particularly if costs go up before they go down.
Costs are viewed like any other resource—capacity, opex for labeling budgets, and payroll for headcount all have to be deployed across teams. In a year or two, the burn rate of a strong engineer might be the same as their salary or cost of employment. In that world, caps will probably need to be put in place. The cap should be proportional to the company's trust in your ability to use tokens in an ROI-positive way. Costs will go up because more tokens will be used, not necessarily because prices will go up, but then prices will come down because frontier models will be in a pricing war.
Taste and judgment are where human brains will continue to be most valuable as AI eats more of the product development life cycle.
AI is effective for providing feedback on strategy but not for generating strategy independently. The user should define goals, vision, constraints, job, and budget, then use AI more like a management tool. This involves defining success, deciding how prescriptive to be about the path to success, and giving feedback along the way. The craft lies in finding the right level of autonomy to give teams or agents. Too much prescription stifles good ideas, while being too open-ended can lead teams in the wrong direction.
Vision is an articulation of the world or state of the product you want to reach. Strategy is an opinionated path to achieve that vision. A strategy must be controversial enough that a reasonable person could disagree with it. Otherwise, you are likely just competing on raw execution. Both vision and strategy will increasingly occupy more of leaders' mental cycles, with less focus on execution.
AI should theoretically excel at strategy given access to market information, competitors, metrics, growth data, and other inputs. In practice, AI produces weak strategies unless steered aggressively with constraints. Effective strategy requires considering technology state, team personnel and motivations, talent attraction potential, competitive landscape, regulatory and compliance landscape, and brand identity. Lazy prompts yield predictable strategies that competitors would expect. High-quality strategies require thoughtful input curation and iterative conversation. Instructing AI to be critical improves results. Different models vary in their willingness to push back; selecting one that challenges assumptions is recommended.
The centaur model has the human in charge with AI assisting, which is preferred. The reverse centaur occurs when AI controls the human, who merely executes its directives. This dynamic is undesirable. People do not want AI dictating strategy and telling them what to build.
Not every workflow should be outsourced to AI even when technically possible. Some tasks represent clear wins, while others carry risks that outweigh benefits.
The strongest product leaders function more as curators than pure visionaries. They curate people, ideas, technologies, and strategies. Success depends on whether an excellent strategy exists and has buy-in, not on its origin. Great curators need some ideas of their own but recognize they cannot generate everything. They create environments where strong ideas surface and get selected. They also curate teams, ensuring leadership groups complement each other in skills and weaknesses while maintaining trust and rapport. Chemistry within leadership teams matters significantly; strong rapport allows teams to navigate most challenges.
A common misconception is that the algorithm possesses detailed semantic understanding of user interests. Progress in recommenders has largely come from large embedding models producing illegible artifacts—giant vectors in high-dimensional space. The algorithm does not semantically understand interests like surfing; it holds numerical correlations. LLMs now enable describing these embedding spaces in natural language. Instagram's "Your Algorithm" feature shows users the topics the system believes they are interested in and allows adding or removing topics. Users can also request non-topical adjustments such as more fun content, more friend content, or constraints on post frequency.
Pure chronological feeds create problematic incentives: everyone posts as much as possible, allowing professional publishers and large companies to dominate because they generate far more volume than individuals. Recency is one input into relevance but not the only one. Algorithmic feeds drive higher engagement than chronological ones. When Instagram made chronological the default, both usage and overall sentiment declined according to large-scale surveys.
AI content represents a tailwind for Instagram, though it creates ranking challenges. In a world of abundant synthetic content, people will increasingly seek creativity, authenticity, and real individuals rather than less. Instagram has long prioritized creators broadly defined—anyone using the platform to advance their work, whether journalists, artists, or small business owners. The platform should not filter content based on the tool used to create it. Content should be evaluated on its own merit, point of view, and the person behind it. Users should know when content is AI-generated and receive information about accounts to make informed trust decisions. Instagram plans to label AI content where detectable, though detection may become harder as models improve. Labeling camera-captured content as non-AI may prove more practical long-term. At the account level, fake AI accounts used for spam require active mitigation.
Examples of high-quality AI content include "Plastic Dream Sequence," featuring doll-like figures in short singing vignettes with a distinct aesthetic, and work by an artist in Paris who combines multiple models to create dreamscapes and animations with a consistent personal style.
Adam Mosseri discusses features from other platforms that he finds impressive, particularly TikTok's recommender system and its ability to surface small creators and break new talent. He explains the difference between exploitation-based ranking, which uses existing data to show content people will likely engage with, and exploration-based ranking, which tests content to discover new interests users may not yet recognize. Exploration-based ranking benefits niche and small creators by giving them opportunities to find audiences they might not otherwise reach.
Instagram has invested heavily in ranking systems over recent years, focusing not only on engagement but also on originality, breakout content, and recency to maintain cultural relevance. Much of this work draws inspiration from TikTok and ByteDance. Mosseri notes that Instagram is ahead in some areas, behind in others, but has a clear path toward becoming best-in-class in recommendations.
Mosseri addresses complaints about the global algorithm not showing friends' content, explaining that this same system enables new creators to reach wide audiences. He emphasizes that most major platform debates involve inherent trade-offs. For example, wanting to see only familiar content leads to lowest-common-denominator experiences, while seeking discovery means occasionally encountering content that misses the mark.
He extends this principle to other contentious issues like privacy versus safety, noting that scanning messages creates significant trade-offs on both sides. Mosseri states his goal when engaging in public debates is not to convince those with fixed opinions but to enumerate the different considerations for others watching the conversation.
Mosseri explains how he became the public spokesperson for Instagram's decisions. The approach began during his time running News Feed at Facebook, when he decided the debate would happen regardless and it was better to participate. He started posting actively on Twitter, where journalists were active at the time, believing it showed humility to engage on their platform.
He describes how Meta's importance in daily life brings responsibility, change, anxiety, stress, and scrutiny. The company has made both great decisions and mistakes, and has faced both fair and unfair criticism. Mosseri believes it is better to communicate openly about decisions, the reasoning behind them, and the associated trade-offs, even if it does not always win friends. He notes that people are more afraid of things they do not understand or that seem secretive.
A senior leadership debate also contributed to this role, where the team discussed moving away from conventional press releases toward using their own platform for communication. Mark Zuckerberg encouraged Mosseri to test this approach.
Mosseri shares strategies for managing hate and criticism. He recounts launching a News Feed redesign in March 2009 as an entry-level designer, receiving a derogatory comment that left him devastated. He developed perspective by recognizing that users organize their digital lives on the platform and would reasonably be upset by sudden, unexplained changes.
He recommends stepping away from online criticism by spending time with family and outdoors. Some months the criticism is manageable, while others it is more difficult. He also describes the redesign of the feed into a video-focused experience that generated significant protest. Multiple changes, including a video viewer test, increased focus on Reels, more recommended posts from accounts users do not follow, and creator complaints about reach, became conflated in public perception.
The controversial video viewer redesign was only an early test to 4% of iOS users and was never intended to launch. Despite the backlash, Instagram continued investing in video, creative tools, ranking, and recommendations, which drove most of the platform's growth in subsequent years.
Mosseri explains that testing at Instagram's scale requires proactive communication planning. For any potentially controversial design change or test, the team discusses messaging in advance, considering both proactive and reactive approaches. He notes that companies cannot launch features to three billion people without testing, but testing inevitably generates coverage, requiring readiness to respond before knowing whether a launch will occur.
He references an incident where Anthropic's head of growth launched a pricing experiment that generated significant Twitter attention, noting that pricing changes require particular care.
Mosseri describes two significant failures. The first was Facebook Home, his initial project as a product manager, which was a fork of Android at the operating system level paired with HTC hardware. The project was a spectacular failure, but he learned extensively about carriers, OEMs, certification, Android, and operating systems. He considers the experience one of the most educational periods of his career and values having brought the project to an end rather than continuing without market fit.
The second failure involved the first version of Reels, built on top of Stories in 2019. Stories had strong momentum, but the approach proved flawed because Stories have relatively low read-through rates and most Reels were never seen before disappearing. Mosseri believes launching a stronger Reels version in summer 2019 rather than summer 2020 would have positioned Instagram better when TikTok grew significantly during the pandemic. He acknowledges that extending existing primitives rather than adding new ones was a design philosophy that proved costly for the business.
Mosseri outlines his approach to screen time with his three children, ages 10, 8, and 6. The core principle is establishing boundaries rather than allowing on-demand access. Children earn iPad time through activities such as completing homework, with 90 minutes available on weekends. Parents should approve which apps children can download.
Exceptions include long flights where devices help manage travel. He supports no-phone policies in classrooms. Regarding AI, he balances concerns about children losing critical thinking skills against the risk of falling behind in AI literacy. He has begun "vibe coding" with his 10-year-old, building a 19-level platformer game resembling 8-bit Super Mario, complete with themed levels, monsters, and a store for skins and weapons. The goal is teaching children to create rather than only consume, while maintaining supervision.
Mosseri emphasizes that every child is different, but digital and AI literacy are important to avoid disadvantage. The approach requires boundaries, scoping activities to healthy options, and avoiding a free-for-all.
In closing, Mosseri encourages listeners to remember that technology and platform decisions are complicated and almost always involve trade-offs. He invites criticism and feedback but asks that people recognize these debates are rarely as simple as they are often portrayed, and that the teams making decisions are people trying to do their best.
Keep Lenny's Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
