All podcasts / Lenny's Podcast / Summary

The rise of taste, human authenticity and judgment in an AI world | Adam Mosseri (Head of IG)

2026-07-09 - 68 min - source - Read full transcript
Lenny Rachitsky (host)Adam Mosseri

Key insights

Taste and judgment become more valuable, not less, as AI makes building cheap.
Mosseri opens by arguing that in a world where it's easy to build almost anything, the scarce skill shifts to knowing what's worth building in the first place. He says the people who benefit most from AI will be clear-eyed about what it's currently good and bad at, and have an instinct for where that line is moving next, rather than treating AI capability as a binary.
taste-and-judgment
Meta has replaced its baker's-dozen specialist team structure with small generalist pods built around a new 'product staff' role.
The old default was roughly 2-3 engineers per platform plus a dedicated PM, designer, data scientist, and researcher - on the order of 13 people. In 2026 Meta shifted to 4-6-person pods, often just four engineers and one product staff generalist who absorbs PM, design, and data-science work using internal AI tools, with senior specialists pulled in only when the work genuinely requires deep expertise (e.g., a pricing algorithm).
product-org-design
Mosseri is bullish on designers specifically because taste is hard to automate, even as functional lines blur.
He expects many of the strongest future product-staff generalists to be converts from design and data science, since designers in particular tend to carry strong opinions beyond visual craft - on strategy, business, and go-to-market - that transfer well once the boundaries between roles dissolve. He flags this as a partial bias given his own background as a Facebook designer.
taste-and-judgment
His baseline hiring bar hasn't changed, but the marginal premium has shifted to curiosity and a willingness to look foolish trying new tools.
He has always screened for grit, being a quick learner, and self-awareness. What's newly important, he says, is behaving like someone learning a new language: willing to sound like an idiot, be corrected, and keep trying rather than waiting to be competent before acting. He sees this as the main predictor of who adapts well as tools and models keep shifting.
taste-and-judgment
Meta rejected leaderboards for AI token spend and has no per-engineer token caps yet, but expects trust-based caps once AI burn rate approaches an engineer's salary.
He treats tokens as just another resource to allocate, like GPUs, storage, or headcount budget, and says token leaderboards created bad incentives without producing value ('it's not that hard to build a token incinerator'). He expects token costs to rise before falling due to a coming frontier-model pricing war, and predicts spend caps will eventually be set proportional to how much a company trusts a given team to spend ROI-positively.
product-org-design
Human judgment is concentrating in vision and strategy, not execution, and AI strategy output is mediocre unless heavily steered.
Mosseri distinguishes vision (an articulation of the desired end state) from strategy (a debatable, opinionated path to it - if nobody could reasonably disagree with it, it's not really a strategy). He says asking an AI for strategy 'lazily' produces the same predictable answer a competitor would expect; getting something genuinely useful requires explicitly feeding it the full set of real-world constraints - team composition, competitive and regulatory landscape, brand identity - and picking a model willing to push back rather than please.
taste-and-judgment
AI content abundance is a tailwind for Instagram because scarcity of authentic, identifiable creators becomes more valuable, not less.
He frames this in terms of a longer-running shift of power from institutions to individuals (comparing it to sports, where players now outdraw teams). As synthetic content floods every platform, he expects people to actively seek out creativity and a knowable point of view behind the content, and argues Instagram's advantage is that it is already the largest creator platform by his internal definition.
ai-content-and-authenticity
Instagram will label content and accounts by AI origin but explicitly refuses to rank or filter based on that label.
Mosseri says the company will try to tell users whether content is AI-made or an account is likely AI-run (noting detection will get harder as models improve, so confidence levels matter), and may find it more practical long-term to label verified camera-captured content rather than label AI content. But he insists content should be judged on its point of view and the person behind it, not on the tool used to make it - separate from cracking down on AI-run spam accounts impersonating real, trustworthy sources.
ai-content-and-authenticity
Instagram's ranking is built on illegible embedding-space math, not semantic understanding, and LLMs are only now making it human-readable.
He says the common assumption that the algorithm 'knows you like surfing' in a legible, semantic way is mostly wrong; the underlying signal is a high-dimensional vector with no inherent meaning to a person. What's new is that an LLM can now describe what a region of that embedding space corresponds to in plain language - the basis for Instagram's 'see your algorithm' feature that lets users view and edit an approximation of their inferred interests.
algorithmic-feeds
Pure chronological feeds fail at scale because they reward posting volume over relevance, and Meta's own tests show it hurts both usage and sentiment even when users say they want it.
A chronological feed incentivizes everyone to post as often as possible since every post always appears at the top, which lets large publishers and companies flood the feed at the expense of friends who post rarely. When Meta made chronological the default in tests, both usage and long-run sentiment dropped even though the choosing individual initially felt more in control, which he attributes to recency being only one input into relevance, not the only one.
algorithmic-feeds
Mosseri treats public criticism as structural and permanent, not something to win, and requires a pre-planned communications response for any test that could leak.
He traces his habit of engaging critics directly back to running Facebook's News Feed, where he decided the public debate would happen with or without Meta's participation. He says he isn't usually trying to change strongly-held minds but to surface the tradeoffs for everyone else watching, and that at Instagram's scale every potentially controversial design test now needs an agreed communications plan for when, not if, it leaks.
leading-through-controversy
His two biggest professional failures were Facebook Home and building the first version of Reels on top of Stories instead of as its own product.
Facebook Home, his first PM project, was a failed Android-fork-plus-HTC-hardware bet; he says shipping and killing it was still valuable because it definitively tested the idea. More costly was building early Reels inside Stories in 2019 to piggyback on its momentum - Stories' low read-through rate meant most Reels went unseen, leaving Instagram poorly positioned when TikTok's pandemic-era breakout hit in 2020, a decision he calls a real fork in the road for the business.
leading-through-controversy

Media referenced

Companies

Techniques and frameworks

Summary

Adam Mosseri, head of Instagram for the past eight years and previously the architect of Facebook's News Feed and ranking algorithm, joins Lenny to argue that taste and human judgment become more valuable, not less, as AI collapses the cost of building things. The conversation opens on how this plays out inside Meta's own org design: 2026 has brought a shift from "baker's dozen" specialist teams (dedicated engineers per platform plus a PM, designer, data scientist, and researcher) to small generalist "pods" of four to six people, anchored by a new "product staff" role that folds PM, design, and data-science work into one generalist who leans on internal AI tools to do what used to require a specialist. Mosseri is explicit that this doesn't eliminate specialists - it just concentrates them on the work that actually requires deep expertise - and he says he remains bullish on designers specifically, since taste is hard to automate and many of Instagram's strongest future generalists are likely to be design and data-science converts rather than PMs.

On hiring and career advice, Mosseri says his baseline bar hasn't moved (grit, being a quick learner, self-awareness), but the new premium is curiosity paired with a willingness to look foolish trying unfamiliar tools, which he compares to the discomfort of learning a new language. He's openly skeptical of using AI for strategy work by default: asking a model for a strategy "lazily" produces the same predictable answer a competitor would expect, and getting something genuinely useful requires steering it with real constraints (team, competitive landscape, regulatory environment, brand identity) and picking a model willing to push back rather than flatter. He frames the durable human contribution as vision (articulating a desired end state) and strategy (a debatable, opinionated path to it, one a reasonable person could disagree with) rather than execution, which he expects AI to keep absorbing.

A significant stretch covers how Instagram thinks about AI-generated content. Mosseri argues the rise of synthetic content is a net tailwind for Instagram, reasoning that as content becomes abundant and cheap, people will increasingly seek out identifiable, authentic creators with a real point of view - and that Instagram's scale as a creator platform positions it well for that shift. The company plans to label content and accounts by AI origin (acknowledging detection will get harder as models improve) but explicitly refuses to rank or filter content based on whether AI made it, judging it instead on its point of view and the person behind it - while still cracking down on AI-run spam accounts that impersonate trustworthy sources. He also unpacks a common misconception about the Instagram algorithm: it has historically run on illegible embedding-space math rather than legible semantic understanding of user interests, and it's only now that LLMs can translate those opaque vectors into plain-language descriptions, which is the basis for Instagram's new "see your algorithm" feature.

The conversation closes on leadership under public scrutiny. Mosseri traces his habit of directly engaging critics back to running Facebook's News Feed, where he concluded the debate over changes would happen whether or not Meta participated, and now treats every design test with controversy potential as requiring a pre-agreed communications plan for when it leaks, not if. He names two career failures candidly: Facebook Home, his first PM project and a doomed Android-fork-and-hardware bet with HTC, and building the first version of Reels on top of Stories in 2019 rather than as its own product, which left Instagram poorly positioned when TikTok's pandemic-era breakout hit in 2020. He ends on a personal note about setting screen-time boundaries with his three kids and vibe-coding a platformer game with his 10-year-old in Claude Code, and closes with a message to critics: the tradeoffs behind Instagram's decisions are almost never as simple as public debate makes them look.

Notable Quotes

"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." - Adam Mosseri

"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." - Adam Mosseri

"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." - Adam Mosseri

"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, just trying to do the best we can." - Adam Mosseri