Key Moments

Adam Mosseri: Building Instagram for an AI world

Lenny's PodcastLenny's Podcast
People & Blogs5 min read69 min video
Jul 9, 2026|2,463 views|77|5
Save to Pod

Want to know something specific about what's covered?

We've already dissected every moment. Ask and we will deliver (with timestamps).

TL;DR

AI will be a tailwind for Instagram, but people will seek out authenticity more than ever in a sea of synthetic content. The platform must invest in creators to stay relevant.

Key Insights

1

Instagram's algorithm doesn't possess a detailed semantic understanding of user interests; instead, it relies on complex, unreadable 'embedding' vectors that correlate with preferences.

2

Product teams at Meta are shifting to smaller 'pods' of 4-6 generalist engineers and a 'product staff' role (an evolution of PMs) augmented by AI tools, reducing the need for traditional specialists.

3

Taste is a crucial skill that AI struggles to replicate, making designers particularly valuable, even as their roles evolve into more generalist product staff positions.

4

The rise of AI content is seen as a tailwind for Instagram, as people will increasingly seek out authentic creators and unique points of view amidst an abundance of synthetic content.

5

Meta is exploring labeling AI-generated content and even potentially camera-captured content, aiming to provide users with more transparency about the origin of content and accounts.

6

Adam Mosseri believes that while executing tasks may become more automated, human brains will remain most valuable for defining vision, strategy, and judgment, especially within defined constraints.

The evolving product team structure

Instagram's head, Adam Mosseri, discusses a significant shift in product team structure at Meta, moving from larger, specialized teams to smaller 'pods' of 4-6 engineers. These pods are supported by a 'product staff' role, an evolution of the Product Manager that leverages AI tools to perform some design, data science, and research tasks. This structure aims to increase speed and reduce 'design by committee.' While specialists are still needed for novel or complex tasks, the core team is leaner, allowing for more agile decision-making. The rationale behind this shift is not solely AI-driven productivity gains but also the inherent effectiveness of smaller, more cohesive teams.

The enduring value of taste and human judgment

In a world where AI makes building things easier, Mosseri emphasizes the amplified importance of 'taste' – the ability to discern what should be built. This is particularly relevant for designers, whose roles are evolving but whose core skill of taste is difficult for AI to automate. He notes that designers have strong opinions that extend beyond interaction and visual design into product strategy and go-to-market, making them prime candidates for the new product staff roles. While many functional lines are blurring, exceptional designers are likely to thrive by leveraging their taste and expanding their influence across broader product decisions.

AI as a tailwind, but authenticity is key

Mosseri views the rise of AI-generated content as a potential tailwind for Instagram, not a headwind. He reasons that in an era flooded with synthetic content, users will increasingly seek out creativity, authenticity, and human connection. For Instagram, which has long prioritized creators, this trend means leaning further into supporting individuals using the platform to share their unique perspectives. While acknowledging the challenge of ranking AI content and the potential for spam, Instagram's strategy is to embrace AI content but clearly label it, allowing users to make informed decisions based on the creator and their point of view, rather than the tool used for creation.

Reimagining algorithmic understanding of users

A common misconception is that Instagram's algorithm has a deep, semantic understanding of user preferences. In reality, for years, recommendation systems relied on complex, unreadable 'embedding' vectors—essentially large numbers that correlated with interests. Mosseri explains that LLMs are now enabling a more interpretable understanding, allowing the system to describe user interests in natural language, like 'deep pour-over coffee snobbery.' This advancement is crucial for initiatives like 'your algorithm,' which aims to give users more agency by allowing them to see and adjust their interest profiles, moving beyond purely topical recommendations to broader desires like 'more fun content' or 'seeing friends more.'

The future of product development: Vision and strategy over execution

As AI increasingly automates execution-level tasks in the product development lifecycle, Mosseri believes human brains will remain most valuable in defining vision and strategy. This involves setting clear goals, articulating a desired state, and forming opinionated paths to achieve it, acknowledging that these strategies must be debatable to avoid simply competing on execution. He likens leadership to management, balancing prescription with autonomy to prevent stifled ideas or wasted time. While AI can provide input and analysis, the core responsibility for vision, strategic direction, and judgment—especially within the complex interplay of technology, team, market, and brand identity—will remain a distinctly human domain.

Navigating content labeling and AI spam

Instagram is grappling with how to label AI-generated content. While detecting AI content is currently possible, Mosseri anticipates this will become harder as models improve. He suggests that labeling camera-captured content as 'non-AI' might be more practical long-term. Beyond content, there's a concern about AI being used for new spam vectors, such as fake accounts posing as AI creators to push illicit products. The platform aims to ensure users know if content or accounts are AI-generated and whether accounts are new or established, empowering them to make informed decisions.

Learning from failures: Facebook Home and early Reels

Mosseri shares two significant failures that provided valuable lessons. Facebook Home, an ambitious OS-level fork of Android for hardware, was a spectacular failure that significantly accelerated his learning curve as a new Product Manager, teaching him about carrier and OEM relationships. Another misstep was building the first version of Reels on top of Stories, which had momentum but was not a suitable foundation. Because Stories content often disappeared quickly, many Reels were never seen. This out-of-position strategy contrasted with TikTok's explosion during the pandemic, highlighting the critical importance of timing and underlying platform architecture for new features.

Screen time, digital literacy, and AI for kids

For his three young children (10, 8, and 6), Mosseri prioritizes boundaries and education regarding screen time. Children must earn their device time, and he personally approves all apps. He advocates for this approach at a policy level. He acknowledges the valid concern about AI hindering critical thinking but also worries about children being disadvantaged if they don't learn to leverage AI. Experimenting with his eldest, they engage in 'vibe coding' together to build video games, focusing on creation and learning AI tools. The goal is to foster digital and AI literacy, ensuring children can use these technologies effectively and responsibly without a 'free-for-all' approach.

Common Questions

Contrary to popular belief, the Instagram algorithm doesn't have a highly detailed semantic understanding of user interests. Historically, it relied on 'embedding models' that create complex mathematical representations (vectors) correlating with interests, rather than understanding them in human terms like 'surfing'. However, with the advent of LLMs, Instagram is beginning to describe these interests more semantically, allowing users to see and adjust their algorithm.

Topics

Mentioned in this video

More from Lenny's Podcast

View all 40 summaries

Ask anything from this episode.

Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.

Get Started Free