LLenny's Podcast
← All episodes
09 July 2026

Adam Mosseri: AI is a tailwind for authenticity

7Frameworks
15Insights

Frameworks in this episode

Innovation3 steps

Steering AI to a Non-Obvious Strategy

AI gives predictable strategy when asked lazily — enumerate every input first, then make it argue back

Read →
Leadership3 steps

The Curator Leader (Not the Visionary)

The best product leaders don't own the ideas — they build environments where great ideas bubble up and get chosen

Read →
Leadership3 steps

The Pod + Product Staff Team Model

Replace the 13-person specialist team with a 6-person generalist pod plus one summoned specialist

Read →
Leadership3 steps

The Three-Trait Hire Plus Two AI-Era Premiums

Grit, quick-learning, self-awareness get you hired anywhere — curiosity and putting yourself out there keep you relevant

Read →
Finance3 steps

Token Budgets as a Trust-Proportional Resource

Kill the token-spend leaderboard; cap AI spend like any resource, sized to trust in someone's ROI judgment

Read →
Strategy3 steps

Vision-Strategy-Execution Split (and the Controversy Test)

Vision is the destination, strategy is an opinionated path — and a real strategy must be disagreeable

Read →
Communication3 steps

When-It-Leaks Communications Pre-Mortem

At scale you can't test quietly — decide the message before you even know you want to launch

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Myth Buster· 2

Myth Buster18:30

Stop treating AI as binary — the tools are great at some things, bad at others

Mosseri rejects the polarized 'AI pill vs anti-AI' framing. He says the tools themselves aren't binary either: they're amazing at some tasks and remarkably bad at others. The people who win are the ones who are clear-eyed about where AI is strong and weak today, and who have a nose for where it's headed. He also notes the underlying job has changed — engineers now spend most of their time planning and reviewing code, not writing it.

  • The 'are you AI pill or anti-AI' debate is a false binary; people aren't binary and neither are the tools
  • AI is amazing at some things and remarkably bad at others
  • Winners are clear-eyed about current strengths/weaknesses and can anticipate future ones
  • Engineering used to be 40-60% writing code; at the labs it's now mostly planning and reviewing
  • Who succeeds depends on whether your strengths align with the tools' and the business's needs

“really polarized binary outlook on the state of AI, like are you AI pill or are you anti- AI? It's like people aren't binary.”

Adam Mosseri · 19:00

“they're amazing at some things and remarkably bad at others”

Adam Mosseri · 19:00
#ai#engineering#future of work
Myth Buster34:30

The Instagram algorithm knows far less about you than people assume

The most common misconception, Mosseri says, runs opposite to the fear: people assume a detailed semantic understanding of their interests that doesn't exist. Recommender progress came from large embedding models that produce illegible giant vectors, not human-readable interests. The system doesn't 'know' you like surfing — it has a number that happens to correlate with surfing. Only recently, via LLMs, is Instagram getting as sophisticated as people already assumed it was.

  • People overestimate how detailed the algorithm's semantic understanding of them is
  • Recommender gains came from embedding models producing illegible vectors, not readable interests
  • The system doesn't know you like surfing; it has a number that correlates with surfing
  • Historically it was collaborative-filtering-style: people who liked these photos also liked those
  • Only recently, with LLMs, is Instagram becoming as sophisticated as users long assumed

“It just has this big ass number that happens to correlate with surfing.”

Adam Mosseri · 35:00

“a misconception historically is until recently, we don't really know as much about you as you think”

Adam Mosseri · 37:30
#algorithm#recommenders#instagram#ai

Hot Take· 5

Hot Take08:00

Why Mosseri is 'long on designers' even as design jobs feel threatened

With generalists, engineers, and product staff all now doing design work, many designers are anxious about their roles. Mosseri pushes back: because designers tend to have taste, and taste is the hardest thing to imagine being automated, he's optimistic about them. He argues the strongest designers won't stay in traditional design roles and will convert into product staff who carry opinions across strategy, business, and go-to-market.

  • When building is easy, the scarce skill becomes deciding what to build in the first place
  • Taste is much harder to automate than execution, which is why he's bullish on designers
  • Despite the optimism, designer job postings are flatlining; the missing piece is business/PM judgment
  • The strongest designers will migrate into product-staff roles, keeping deep craft plus cross-functional opinions

“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…”

Adam Mosseri · 08:00

“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…”

Adam Mosseri · 08:30
#design#taste#ai#careers
Hot Take21:00

'No leaderboards for token spend' — how Mosseri thinks about AI budgets

Meta was famous for a token-spend leaderboard, which Mosseri calls a terrible idea. Instagram reined in costs by shutting down low-value 'token incinerator' projects rather than capping engineers. He predicts that within a year or two a strong engineer's AI burn rate could match their salary, at which point proportional caps tied to ROI trust will become healthy — but for now there are no limits.

  • Meta's token-spend leaderboard was a bad idea; competing on spend creates no value
  • It's easy to build a 'token incinerator' that burns money without creating value
  • Instagram currently has no token limits for engineers or most people
  • In a year or two an engineer's AI burn rate may equal their cost of employment
  • Caps should be proportional to the company's trust in your ROI-positive use of them
  • He expects costs to rise (more tokens used) then fall as frontier models enter a price war

“it's a terrible idea. No leaderboards for token spend.”

Adam Mosseri · 21:00

“the burn rate of a strong engineer might be the same as their salary or their cost of employment”

Adam Mosseri · 22:30
#ai#budgets#engineering#meta
Hot Take30:30

The best product leaders are curators, not idea machines

Mosseri argues the strongest product leaders are less prolific visionaries and more curators — of people, ideas, technologies, and strategies. He doesn't care whether a great strategy came from a lead or someone else, only that it's amazing, bought into, and executed well. Great curators still need their own ideas, but they build environments where great ideas bubble up and get chosen.

  • The best leaders are curators of people, ideas, technologies, and strategies
  • It doesn't matter whose idea the strategy is, only that it's great and everyone's bought in
  • You can't be a great curator with no ideas of your own
  • The job is creating an environment where great ideas surface and get decided upon
  • Curating teams and chemistry matters as much as curating ideas

“curators of people, curators of ideas, curators of technologies, curators of strategies”

Adam Mosseri · 31:00

“I just care that there is a amazing strategy and everyone has bought into it”

Adam Mosseri · 31:30
#leadership#product#management
Hot Take41:00

AI content is a tailwind for Instagram — because people will seek authenticity

Asked whether AI content is a headwind or tailwind, Mosseri says tailwind, but a challenge. As an attention business, more content is potentially more attention, but Instagram isn't yet good at ranking AI content. His deeper bet: as power shifts from institutions to individuals (like players over teams in sports), and as synthetic content floods in, people will seek out creativity, authenticity, and real people more, not less — which favors the largest creator platform.

  • AI content is a tailwind but a challenge; ranking AI content well isn't solved yet
  • Power is shifting from institutions to individuals across industries (e.g. athletes over teams)
  • Instagram is leaning into creators broadly, not just branded-content influencers
  • In a flood of synthetic content, people will seek out authenticity and real people more
  • Instagram was never just about content but about the person and point of view behind it

“I think it's going to be a tailwind, but I think it's going to be a challenge.”

Adam Mosseri · 41:00

“I actually think people are going to seek out creativity and authenticity and people [snorts] more, not less.”

Adam Mosseri · 42:30
#ai#creators#authenticity#instagram
Hot Take43:30

Don't judge content by the tool that made it — label it instead

Mosseri argues Instagram shouldn't filter out AI content or make value judgments based on the tool used; it should judge content on its point of view and the person behind it, and simply let users know whether something is AI. Detection will get harder as models improve, so honesty about confidence matters. He suggests it may be more practical long-term to label camera-captured content, and to mark accounts (real person vs not) to fight AI-driven spam and scams.

  • Don't judge content by the tool that made it; judge the content, point of view, and person
  • Don't filter out AI content — let users know whether content is AI
  • Detection may weaken as models improve, so be honest about confidence levels
  • Labeling camera-captured (non-AI) content may be more practical long-term than labeling AI
  • Marking accounts matters to fight spam vectors like fake 'AI monk' supplement scams

“I don't think we should judge content based on the tool that made it. Um I think we should judge it based on the content,…”

Adam Mosseri · 44:00

“I think we should let you know if content is AI content or not.”

Adam Mosseri · 44:30
#ai#content moderation#labeling#instagram

Explainer· 4

Explainer02:30

Instagram's shift to small 'pods' and the 'product staff' role

Mosseri describes how Instagram replaced the traditional ~13-person team (specialists across iOS, Android, server, PM, design, data science, research) with 'pods' of four to six generalist engineers plus a 'product staff' generalist and whatever senior specialist the work requires. He frames the smaller core as the big organizational change of the year, arguing productivity gains come as much from small teams as from AI itself.

  • The old canonical team was ~13 people heavy on specialists; pods are ~6-7 with a smaller core
  • 'Product staff' is an evolved PM who can do some design, data science, and research work
  • Teams pull in a senior specialist only when the work demands it (e.g. pricing needs a senior data scientist)
  • Fewer people to coordinate means faster moves and less design-by-committee
  • Gains come partly from AI tooling and partly just from small teams being more effective

“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”

Adam Mosseri · 03:00

“they just by virtue of having less people to coordinate they can often move faster and make um better decisions a little bit less design…”

Adam Mosseri · 04:00
#product teams#org design#ai#management
Explainer25:30

Why AI is bad at strategy unless you steer it hard on constraints

You'd expect AI to be great at strategy given all the market data, but Mosseri finds it isn't unless steered aggressively — not toward an answer, but toward the constraints. Good strategy requires weighing technology, team and talent, competitive, regulatory, compliance, and brand-identity factors. Ask lazily and you get predictable output the competition would expect; put in the work and pick a model willing to push back, and it becomes genuinely clarifying.

  • AI isn't good at strategy unless you steer it aggressively on constraints, not toward an answer
  • Strategy inputs include technology state, personnel/motivation, competitive, regulatory, compliance, and brand identity
  • A lazy prompt yields predictable output competitors would expect
  • Enumerate the inputs, make it a back-and-forth conversation, and tell it to be critical
  • Different models have different willingness to push back; pick one that does

“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.”

Adam Mosseri · 26:00

“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.”

Adam Mosseri · 27:00
#strategy#ai#prompting#product
Explainer35:30

'See your algorithm': using LLMs to give users agency over their feed

Mosseri describes a feature where Instagram maps everything you've interacted with into an embedding space and has an LLM describe regions of that map in plain language (e.g. 'deep pour-over coffee snobbery'). This lets users see the topics the system thinks they're interested in and add or remove them. The goal is giving people agency back in a world increasingly run by recommendations, with future controls beyond topics ('show me my friends more', 'no seven photos in a row').

  • Your interactions are mapped into an embedding space; an LLM describes each region in words
  • Users can see the topics the system thinks they like and add or remove them
  • The aim is restoring agency as apps get taken over by recommendations
  • Future, non-topical controls could include 'more fun content' or 'see my friends more'

“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…”

Adam Mosseri · 36:00

“the idea here giving people some agency back in a world where, you know, these social media apps are getting taken over by recommendations”

Adam Mosseri · 37:00
#algorithm#instagram#ai#product
Explainer38:30

Why a pure chronological feed backfires at scale

People believe they'd love a chronological feed of everyone they follow, but Mosseri says it keeps getting proven wrong. A pure chronological feed incentivizes everyone to post as much as possible, so the feed fills with high-volume professional publishers while a close friend's one weekly post drowns. Recency matters but isn't the only input to relevance. When Instagram makes chronological the default, usage and overall sentiment both drop.

  • Chronological feeds reward posting volume, so publishers dominate and friends drown out
  • Designing feeds is like designing a city: consider incentives and how people act within them
  • Recency is an important input into relevance but not the only one
  • When chronological is the default, both usage and long-run sentiment go down in surveys

“the New York Times can pump out 50 things a day. your your best friend won't you might get one thing a week from them…”

Adam Mosseri · 39:00

“Recency is an important input into relevance, but it's not the only one.”

Adam Mosseri · 39:30
#algorithm#feed#instagram#product

Story· 1

Story1:00:30

Two failures Mosseri learned most from: Facebook Home and reels-on-stories

In 'fail corner,' Mosseri names two. Facebook Home, an Android fork with HTC hardware, was a spectacular failure that taught him more than any other year. Bigger, though: the first version of reels was built on top of stories to ride its momentum, but stories was a weak foundation with low read-through, so most reels were never seen. That mistiming left Instagram out of position when TikTok exploded in 2020 — a pretty big fork in the road for the business.

  • Facebook Home (an Android fork plus HTC hardware) was a spectacular failure but a huge learning year
  • Sometimes the best thing is to execute a no-market-fit idea well just to prove it was wrong
  • The first version of reels (2019) was built on stories to chase momentum, but stories was a weak base
  • Low story read-through meant most reels went unseen and quickly disappeared
  • Being out of position let TikTok explode during the 2020 pandemic — a big business fork

“the first version of reels was built on top of stories. Stories had a ton of momentum.”

Adam Mosseri · 1:01:30

“On the other hand, I was wrong and and it's a pretty big fork in the road if you just look at the overall business…”

Adam Mosseri · 1:02:30
#failure#instagram#reels#product

Takeaway· 3

Takeaway14:00

The three traits Mosseri always hires for, plus what matters more now

Regardless of function, Mosseri screens for three things: grit, being a quick learner, and self-awareness. If a candidate has all three they can get good at almost anything; if any is missing there's usually a problem. For the next five to ten years he adds two premiums: staying curious and being willing to put yourself out there and look foolish while you learn.

  • Baseline traits: grit ('fire in your belly'), quick learner, self-aware enough to take feedback
  • With those three, you can eventually get good at anything; missing one signals trouble
  • New premiums for a changing world: stay curious and be willing to try things
  • Learning a new language analogy: the best predictor is willingness to sound like an idiot and be corrected
  • He also expects fewer roles for large-organization managers as teams shrink

“if you got fire in your belly, you learn quickly, and you're self-aware, you can kind of get good at anything eventually”

Adam Mosseri · 14:30

“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…”

Adam Mosseri · 15:30
#hiring#careers#leadership#ai
Takeaway23:30

Where human brains stay most valuable: taste, judgment, vision and strategy

Asked where humans stay valuable as AI eats the product lifecycle, Mosseri points to taste and judgment, especially around strategy. He defines vision as an articulation of the state of the product you want to reach, and strategy as an opinionated path to get there. Working with AI, he says, looks more like management: setting the goal, deciding how prescriptive to be, and giving feedback along the way.

  • You might get feedback from AI on a strategy, but you're not asking it to author one anytime soon
  • Vision = an articulation of the state of the product you want to get to
  • Strategy = an opinionated path to achieve that vision, and it should be controversial enough to disagree with
  • Directing AI resembles management: define success, choose how prescriptive to be, give feedback
  • Too prescriptive stifles good ideas; too open-ended wastes time — the same tension may apply to future agents

“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”

Adam Mosseri · 24:00

“I think of vision as an articulation of the world or the or the state of the product you want to get to. Then I…”

Adam Mosseri · 25:00
#strategy#ai#product#leadership
Takeaway1:03:00

How the head of Instagram handles screen time — and vibe codes with his kids

Mosseri's core principle for his three kids (10, 8, 6) is boundaries plus education and conversation. They're too young for social media, but each earns iPad time (e.g. homework blocks buy weekend minutes), and he approves which apps they download. He's experimenting with AI literacy by vibe coding with his 10-year-old, who built a 19-level platformer — because he's more worried about kids being unable to leverage AI than about the risks, as long as it isn't a free-for-all.

  • Core principle is boundaries, plus education and ongoing conversation
  • Kids are too young for social media but earn screen time (homework blocks buy weekend minutes)
  • Parents should approve which apps kids download; Meta advocates this at a policy level
  • He worries kids not learning to leverage AI will be at a disadvantage — a balance against over-use
  • He vibe codes with his 10-year-old, who built a 19-level Super-Mario-style platformer using Claude Code

“The key thing for me is boundaries.”

Adam Mosseri · 1:03:00

“he's made this 19 level platformer game that kind of looks like an eight-bit version of Super Mario”

Adam Mosseri · 1:05:00
#parenting#screen time#ai literacy#kids