Adam Mosseri: AI is a tailwind for authenticity

Adam Mosseri (guest); Lenny Rachitsky (host) · podcast · 2026-07-11 Reader 🖥️

practical · mixed — A long-form podcast conversation (Lenny's Podcast) between host Lenny Rachitsky and Adam Mosseri (Head of Instagram). Primarily practical: it argues how product builders and leaders should orient themselves and organize teams in an AI era. It mixes argumentative segments (taste/judgment/authenticity as the durable human edge, AI content as a tailwind, small generalist pods), expository segments (how Instagram's recommender actually works, why chronological feeds fail at scale, token-budget management), and narrative segments (Facebook Home failure, reels-built-on-stories mistake, the 2009 Newsfeed redesign and 2022 feed-test backlash, his kids' screen-time policy and a 10-year-old vibe coding with Claude Code). NB: source is auto-generated YouTube captions; several proper nouns were phonetically mangled and have been corrected (Mosseri, Systrom, Krieger, Kevin Weil, Cory Doctorow, Boris Cherny, Claude Code).

Structure

Key terms

• product staff
  An evolution of the PM into a generalist who, leveraging new AI tools, can also do parts of what a designer, data scientist and researcher used to do — the human core of a small 'pod', with senior specialists pulled in only when the work demands it.
• taste Taste & Judgment
  The instinct for what is good and what should be built — the faculty Mosseri thinks is 'much more difficult to imagine being automated away', which makes him 'long on designers' even as generalists encroach on design.
• authenticity as a tailwind
  The claim that an abundance of synthetic/AI content increases (not decreases) demand for creativity, authenticity and 'the person behind the content', so AI content helps Instagram — the largest creator platform — rather than hurting it.
• judgment Taste & Judgment
  Human decision-making around strategy and what to build under constraints — 'you're not asking an AI to come up with a strategy'; the leader defines success, sets constraints and gives feedback, so directing AI 'looks more like management'.
• curator (vs. visionary)
  The best product leaders are less prolific idea-machines than curators — of people, ideas, technologies and strategies — who create an environment where great ideas bubble up and are chosen, and who select for team chemistry (trust and rapport), 'more art than science'.
• reverse centaur Centaur vs. Reverse Centaur
  Cory Doctorow's image (invoked by Lenny) of a horse head on a human body — the AI in charge and the human doing its bidding — the failure mode to avoid, contrasted with the centaur where the human stays in control; used to argue people will not want AI to hand them the strategy.

Propositions

• explicitAs it becomes easier to build things, the scarce and important work shifts to figuring out what you should be building in the first place — so taste matters more, not less.conf: high
  supportsHow Anthropic's product team moves faster than anyone else | Cat Wu (Head of Product, Claude Code) · p3 — Near-identical claim: as building/code becomes cheap, deciding what to build (product taste) becomes the most valuable skill.
• explicitcontestedTaste is 'much more difficult to imagine being automated away', so despite generalists, engineers and product staff now doing design work, Mosseri is 'long on designers' because they tend to have taste.conf: high
  contradictsWhy cultivating agency matters more than cultivating skills in the AI era | Max Schoening (Head of Product, Notion) · p14 — Mosseri claims taste/design is 'much more difficult to imagine being automated away' and is 'long on designers'; Schoening argues taste may NOT be an enduring human advantage because the input-idea-to-reaction loop resembles backpropagation. Direct disagreement on taste's durability.
  supportsAddy Osmani’s Post · p3 — Both name taste as the durable human edge that resists automation.
• explicitThe canonical big-company product team has shifted this year from a 'baker's dozen' of specialists (multiple Android/iOS/server engineers, PM, designer, data scientist, researcher) to a small 'pod' of roughly four-to-six more-generalist engineers plus one 'product staff', with specialists added only when the work requires them.conf: high
  supportsThe engineering manager role is splitting in two · p2 — Both describe AI dissolving the traditional ~7-13-person specialist team; Mosseri's 'product staff' pod is the same collapse of PM/design/data-science glue into a smaller generalist core.
• explicit'Product staff' is an evolution of the PM: a generalist who can do some of what a designer, a data scientist and a researcher do, enabled by new internal AI tools (e.g. automatically pulling a waterfall/funnel analysis that previously required bespoke data-science work).conf: high
• explicitFunctional boundaries are wildly blurring, and some of the strongest future product staff will be converts from design and data science — people who already had strong cross-functional opinions and can now simply jump in.conf: high
  supportsFragments: February 18 · p8 — Both: functional/platform-specialist boundaries are eroding as generalist, cross-craft capability (enabled by LLM tools) becomes more valuable.
• explicitSmaller teams often move faster and make better decisions simply because there are fewer people to coordinate — less design-by-committee — which is a shift driven by more than just AI-boosted productivity.conf: high
  supports9) Nils Brede Moe om AI-paradokset, autonomi, læringskultur og fremtidens team · p3 — Mosseri: smaller teams move faster / less design-by-committee; Moe: the most effective teams are small, fewer people is always better.
• explicitcorroboratedSpecialist functions do not disappear but are reshaped and reduced: you keep a smaller number of senior (or on-their-way-to-senior) specialists and pull in a very senior or more creative specialist precisely when the work needs one.conf: high
  refinesWhy cultivating agency matters more than cultivating skills in the AI era | Max Schoening (Head of Product, Notion) · p20 — Schoening warns role-merging risks losing specialists at the edges; Mosseri supplies the mitigation — keep senior/creative specialists to pull in on demand and deliberately grow tomorrow's seniors.
  supports ←Netflix CPTO on AI and the future of product and tech roles | Elizabeth Stone · p1 — Both accounts agree specialist functions don't disappear under AI-driven role fluidity — they're reshaped and reduced, not eliminated.
• explicitcorroboratedOrganizations must keep investing not only in today's senior specialists but in growing tomorrow's — you cannot rely on super-senior specialists without a pipeline of new ones — or they will regret it in a couple of years.conf: high
  supports ←Netflix CPTO on AI and the future of product and tech roles | Elizabeth Stone · p8 — Both stress that scarce senior/craft expertise must keep being actively grown, not just relied on, even as generalist tooling spreads.
• explicitThe durable hiring baseline is three traits — grit/drive, being a quick learner, and self-awareness (able to take feedback) — because with those three a person 'can kind of get good at anything eventually'.conf: high
• explicitThe traits trending up for success over the next 5-20 years are curiosity and a willingness to put yourself out there and try things (being willing to 'sound like an idiot' and be corrected), precisely because no one really knows what is coming.conf: high
  supportsThe AI paradox: More automation, more humans, more work | Dan Shipper · p10 — Both: curious generalists win the transition because value moves to figuring out what to build; Mosseri elevates curiosity/willingness-to-try as the top trending-up hiring trait.
  supportsDesigning the AI-native engineering organization · p8 — Both put the future skill profile on agency, judgment and generalism over deep specialization.
• explicitLarge-organizational leadership (managing thousands) is trending down as a hiring need, because there will be more smaller teams and fewer roles that manage very large organizations.conf: medium
• explicitAI is resetting who performs: because the job itself has changed (e.g. engineering is now mostly planning and reviewing, not writing code), some former low performers now thrive and vice versa — success is a function of whose strengths align with the tools' and the business's needs.conf: high
• explicitThe state of AI tools is not binary (AI-pill vs anti-AI): the tools are amazing at some things and remarkably bad at others, so the people who make the most of AI are clear-eyed about what it is and isn't good at — and have a nose for what it will become good at.conf: high
• explicitAI spend should be governed like any other resource (GPUs, opex, payroll): public 'leaderboards for token spend' are a bad idea, a 'token incinerator' creates little value, and caps proportional to the company's trust in a person's ROI-positive use will likely become necessary as an engineer's burn rate approaches their cost of employment.conf: high
  supportsThe Agentic P&L: Beyond the Empire of Headcount · p5 — Both treat token/AI cost as a managed resource line — a new permanent P&L/operating constraint requiring active governance (caps, ROI).
• explicitThe cost of intelligence per unit is collapsing: you will still be able to 'write code for free' (with a smaller model), and for side projects a cheap model like Fable will probably be 'just fine', so paying for frontier tokens becomes optional for many tasks even as models diverge in 'vibe'.conf: high
• explicitcontestedHuman brains stay most valuable in taste, judgment, vision and strategy: you don't ask an AI to originate a strategy (only to give feedback within bounds you set), and directing AI 'looks more like management' — defining what success looks like, choosing how prescriptive to be, and giving feedback along the way.conf: high
  supportsWe've Stopped Arguing About Whether AI Changes Software · p9 — Both locate the scarce, valuable human skills in judgment and taste as AI absorbs execution.
  contradicts ←Strategy Making Gets an AI Boost · p3 — Csaszar (via Gavett) casts AI as capable of generating, screening, and effectively architecting strategic alternatives; Mosseri insists you never ask AI to originate a strategy, only to give feedback within human-set bounds — a direct tension on how much strategic-origination work AI should be trusted with.
• explicitAI is not good at strategy unless you steer it hard on the constraints (technology, personnel and motivation, competitive/regulatory/compliance landscape, brand identity): lazy prompting yields predictable answers the competition would expect, whereas real value requires enumerating the inputs, a critical back-and-forth conversation, and a model willing to push back.conf: high
• explicitA good strategy must be controversial — a reasonable person should be able to disagree with it — because a strategy no one could disagree with means you are merely competing on raw execution; vision is the articulated end-state and strategy is the opinionated path to it.conf: high
• explicitThe best product leaders are less visionaries than curators — of people, ideas, technologies and strategies — who care that there is an amazing strategy everyone has bought into rather than that it came from them, and who select for team chemistry (trust and rapport), which is 'more art than science'.conf: high
• explicitcorroboratedAI content is a tailwind (though a challenge) for Instagram versus other platforms: in a world of abundant synthetic content people will seek out creativity, authenticity and people more, not less, and since Instagram is the largest creator platform this benefits it.conf: high
  supports ←Are We Entering a New Age of Creativity with the Help of AI? · p5 — Both argue that abundant AI/synthetic content raises, not lowers, the premium audiences place on trusted human creators and curated interpretation.
  extends ←How a Museum Marketing Team Used AI to Bring People Closer to Art · p2 — Mosseri's general claim that abundant synthetic content drives demand for authenticity/humans is extended here with a concrete mechanism: De Freitas names the specific domain categories (one-of-a-kind originals, high-trust, human-touch) where that authenticity premium is sharpest.
• explicitcorroboratedContent should be judged by its point of view and the person behind it, not by the tool that made it: Instagram should not filter out AI content but should label provenance (possibly labeling camera-captured content rather than AI content long-term) and surface who posted something, letting users make their own informed decisions — while cracking down on deceptive AI spam accounts.conf: high
  supports ←How a Museum Marketing Team Used AI to Bring People Closer to Art · p5 — Mosseri argues content should be judged by provenance transparency rather than concealment; Connolly's categorical commitment to overt AI disclosure (and refusal to use AI covertly in donor communications) is a real-world institutional instance of exactly that provenance-first stance.
  extends ←You Probably Won’t Read This Article…and That’s OK · p6 — Mosseri's answer — judge content by point of view and the person behind it, label provenance, let users decide — is a concrete third gate design (neither pure human-institutional gatekeeping nor an AI reviewer) that this document's two-candidates framing never considers.
• explicitPower has been shifting from institutions to individuals across industries (e.g. athletes are now more relevant than teams), so the strategic bet is to invest in individuals and creators broadly — anyone using the platform as themselves to do what they do.conf: high
• explicitdevelopedPeople overestimate how much the algorithm semantically understands them: modern recommenders run on illegible embedding vectors ('a big ass number that happens to correlate with surfing'), not human-readable interests — and only now can LLMs describe those embeddings in words, enabling features that let users see and adjust 'your algorithm' and reclaim some agency.conf: high
  extends ←You Probably Won’t Read This Article…and That’s OK · p4 — Both critique recommendation algorithms as unreliable arbiters of value — Mosseri from the illegibility of embedding vectors to human interest, this document from the self-reinforcing, winner-take-all popularity dynamic (attention accrues attention).
• explicitEvery major product debate involves genuine trade-offs (algorithmic vs chronological feed, privacy vs safety), so the leader's public role in controversy is usually not to convince opponents but to enumerate the puts-and-takes for everyone else watching, and to be accessible because people fear what is secretive and not understood.conf: medium
• implicitImplicit: the value of a person's work is migrating from producing output to exercising the human faculties AI cannot supply — taste, authenticity, judgment and the ability to steer — so both individual careers and team designs should be reorganized around those faculties.conf: medium

Arguments

Narratives

• Facebook Home — Mosseri's first project as a PM, a fork of Android at the OS level plus HTC hardware — was a spectacular failure, but he learned more in that year-and-a-half than any other year, and he is glad he brought it to an end: sometimes the best thing you can do is execute a no-market-fit idea well enough to decide whether it was a good idea in the first place.
  → Illustrates learning-from-failure and the discipline of killing an idea; supports the 'fail corner' theme and the value of judgment about what to build (and what to stop).
• The first version of Reels (circa 2019) was built on top of Stories because Stories had the most momentum, but Stories was a weak foundation (low readthrough, most reels never seen). By the time the standalone Reels launched (summer 2020) TikTok had exploded during the pandemic; Instagram was 'out of position'. Mosseri, a designer by instinct trying to extend existing primitives rather than add new ones, was wrong — a big fork in the road for the business over eight years.
  → Illustrates the cost of building on the wrong foundation and over-extending existing primitives; a concrete case where taste/judgment about what to build had enormous strategic stakes.
• At the March 2009 Facebook Newsfeed redesign that Mosseri (a 25-year-old IC designer) shipped, the very first comment was derogatory and bigoted and specifically about the design. Devastated, he later reframed it: rearranging millions of people's 'desk' without warning or explanation would reasonably make anyone furious — so he puts backlash in perspective and steps away (time with kids, outside).
  → Illustrates how he metabolizes public criticism and builds empathy for users; supports the leadership-under-scrutiny and accessibility themes.
• The 2022 'protest' over the video-viewer feed redesign was really four things conflated — a 4%-of-iOS test never meant to ship, leaning into Reels, leaning into recommendations, and creators upset about declining reach — amplified by big-name creators publicly slapping Instagram and press coverage echoing back. The takeaway was not that they shouldn't have tested, but that they were pushing a little too fast and that at Instagram's scale you must have a comms strategy ('not if it leaks, when it leaks') before you even know you want to launch.
  → Illustrates that experiments at massive scale are inherently risky and require a pre-planned communication strategy; supports the trade-offs and leadership-at-scale themes.
• Mosseri's screen-time policy for his three kids (10, 8, 6) centers on boundaries: no social media yet, earned iPad time, parent-approved apps. The exception he is experimenting with is vibe coding with his 10-year-old — who built a 19-level eight-bit platformer with its own themes, monsters and item store using Claude Code — because he wants his kids to be digitally and AI literate (learning to make things, not just consume) without a free-for-all.
  → A personal illustration of both the boundary/education philosophy and how accessible building has become — a three-fingered-typing 10-year-old shipping a game in a couple of hours.
• The 'your algorithm' feature: Instagram maps everything a user has interacted with into an embedding space (a map), then has an LLM describe a region of that map in words (e.g. 'deep pourover coffee snobbery'), letting users see and adjust the topics the system thinks they like — an attempt to give people agency back as feeds are taken over by recommendations.
  → Illustrates the claim that recommenders run on illegible embeddings and that LLMs now make them describable, plus the move to give users agency.

Questions addressed

• ✅ What does the canonical product team look like in 2026, and what is 'product staff'?
  A shift from a 'baker's dozen' of specialists to small 'pods' of 4-6 generalist engineers plus a 'product staff' (an evolved PM who absorbs parts of design/data-science/research via AI tools), with senior specialists pulled in only when the work demands.
• ✅ Which adjacent roles (data science, user research, design) are most at risk, and should people shift?
  Anxiety is everywhere and functional lines will keep blurring, but functions won't vanish — they'll be reshaped and fewer; Mosseri is actually 'long on designers' because they carry taste, and expects strong ICs to move fluidly across crafts.
• ✅ What traits are trending up and down in hiring in an AI world?
  Up: curiosity and willingness to try/put yourself out there (on top of the baseline grit, fast learning, self-awareness). Down: large-organizational leadership, because there will be fewer very large teams to manage.
• ✅ Where will human brains continue to be most valuable as AI eats the product-development lifecycle?
  Taste, judgment (especially around strategy), vision and strategy — AI is not asked to originate strategy; steering it resembles management (define success, set constraints, give feedback).
• ✅ How should you manage token spend and budgets for engineers and product teams?
  Treat it like any resource (GPUs, opex, payroll); no leaderboards for token spend; shut down low-value 'token incinerators'; expect caps proportional to trust in ROI-positive use as burn rate approaches salary, even as per-unit cost falls.
• ✅ Is the rise of AI content a headwind or a tailwind for Instagram?
  A tailwind (but a challenge): abundant synthetic content will make people seek out authenticity and people more, and Instagram is the largest creator platform — though it cannot yet rank AI content well and detection may degrade.
• ✅ Should Instagram filter AI content, and how should it treat AI vs. human content/accounts?
  Don't filter by tool — judge content by its point of view and creator; label provenance (possibly labeling camera-captured content long-term) and surface who posted it, while cracking down on deceptive AI spam accounts.
• ✅ What does the Instagram algorithm actually 'know' about users?
  Less than people assume: it runs on illegible embedding vectors, not semantic interests; LLMs now let the system describe those embeddings, powering 'see and adjust your algorithm' features.
• ⭕ Should you outsource a given workflow to AI just because you can?
  Mosseri says no — some cases are win-win, in others the risk outweighs the benefit and humans must stay in control (the reverse-centaur danger) — but he offers no criteria for telling which is which.

Gaps

• incompleteThe tailwind thesis rests on the empirical claim that audiences will reliably prefer authentic/human content over polished synthetic content at scale, but no mechanism or evidence is offered. Mosseri simultaneously concedes Instagram is 'not very good at ranking AI content yet' and that as models improve it 'might lose the ability to detect' AI content — so the very provenance-labeling and ranking that the strategy depends on may be unworkable, leaving the tailwind under-defended.
  Read next: Empirical work on engagement with labeled AI vs. human content and parasocial/authenticity preferences; the AI-slop discourse on how telltale synthetic output degrades information ecosystems; whether 'the person behind the content' is a durable ranking signal.
• incompleteThe 'small pods + product staff' model is drawn entirely from Meta/Instagram — a huge, well-resourced company with a deep bench of senior specialists to pull in on demand. Its viability for startups or teams without that bench is unexamined, and whether generalist product staff can sustain deep specialization, design consistency and accessibility over time is asserted-away ('you still pull in a specialist when you need one') rather than shown.
  Read next: Organizational-design literature on the specialist-vs-generalist trade-off; empirical outcomes from companies that eliminated specialist roles; the same specialization-loss failure mode raised (and left open) in the Catherine Wu / Claude Code discussion.
• incompleteTaste and judgment are presented as durable human advantages 'much more difficult to imagine being automated away', but Mosseri gives no account of why they resist automation, and the opposing view — that taste is an in-group-reaction prediction loop AI can also learn to close — is never engaged.
  Read next: The argument (from the Schoening/Lenny conversation) that taste resembles backpropagation and may not be an enduring human advantage; theories of taste as learnable feedback vs. innate faculty.
• incompleteMosseri says some workflows are win-win to hand to AI while for others 'the risk outweighs the benefits', and warns of the reverse-centaur failure mode, but offers no criteria for deciding which workflows to outsource — leaving the central practical question of human-vs-AI division of labor unresolved.
  Read next: Risk-tiering frameworks for AI-assisted work; human-in-the-loop decision design; the cognitive-surrender literature on when delegation crosses into loss of control.
  closed byEveryone I recommend on AI is optimistic. Except Cory Doctorow. — Partially closes the gap by supplying a portable heuristic question for classifying a workflow as worth handing to AI or not — 'does this let me do better work, or does it just make me keep up with it?' — though this is a single self-applied diagnostic question, not the risk-tiering framework or organizational-level criteria the original gap asked for.

Closes gaps elsewhere

• How Anthropic's product team moves faster than anyone else | Cat Wu (Head of Product, Claude Code) · g2 — Addresses the specialization-loss leg of that gap (not the design-consistency or accessibility legs): Mosseri names loss of deep specialization as the real risk of merging roles and gives Instagram's concrete mitigation — a generalist product-staff core with senior/creative specialists pulled in on demand, keeping designers because they carry taste, and deliberately investing in growing tomorrow's senior specialists so the pull-in model does not starve.

Related documents

• 9) Nils Brede Moe om AI-paradokset, autonomi, læringskultur og fremtidens team
• How Anthropic's product team moves faster than anyone else | Cat Wu (Head of Product, Claude Code)
• What is the harder human-capital problem beneath token capital?
  AI Labor Disruption Coordination Cost Agency Small Teams & Compression
• Hard truths about building in the AI era | Keith Rabois (Khosla Ventures)
  AI Labor Disruption Small Teams & Compression Talent Density Token Throughput
• Hard truths about building in the AI era | Keith Rabois (Khosla Ventures)
  AI Labor Disruption Coordination Cost Small Teams & Compression Talent Density
• Why cultivating agency matters more than cultivating skills in the AI era | Max Schoening (Head of Product, Notion)
  Agency Small Teams & Compression Taste & Judgment
• Designing the AI-native engineering organization
  Agency Small Teams & Compression Taste & Judgment
• The best response to AI slop, infinite advice, and online noise is from Robin Williams
  AI Slop Agency Taste & Judgment
• Presentation: A Few Predicted Talks From QConAI 2030
  AI Labor Disruption AI Unit Economics Token Throughput
• Travel Through the Lens of AI with Booking.com CEO Glenn Fogel
  AI Labor Disruption AI Unit Economics Agency