Adam Mosseri says Meta's canonical product team shrank from about a Baker's dozen people into pods of four to six engineers supported by a blended product staff role.
Mosseri calls Meta's token-spend leaderboard a terrible idea and predicts budget caps may be needed as token costs of strong engineers approach their salaries.
Mosseri argues AI is poor at strategy unless steered hard, and that good strategy must be controversial, while defending algorithmic feeds over chronological ones.
Instagram's Mosseri says AI makes taste and authenticity more valuable
Mosseri argues that as AI makes building cheaper, taste, judgment and clear-eyed use of the tools become the scarce skills. Meta's teams are shrinking into generalist pods while senior specialists still matter. On Instagram, a flood of AI content raises the value of authentic creators, and his answer is labelling and transparency over filtering.
Takeaways
Taste and judgment matter more as building gets easier. When building gets easier, deciding what to build matters more. Mosseri is long on designers because taste is hard to automate. He still wants senior specialists, expects functions to keep blurring, and says companies must keep growing tomorrow's senior talent.
AI content is both tailwind and challenge. Instagram is not yet good at ranking AI content, but creators and individuals will be valued more. Mosseri wants AI content labelled and more shown about the poster, rather than judged by tool. Fake supplement-selling accounts are a new spam vector.
Algorithmic feeds beat chronological ones. A pure chronological feed gets dominated by prolific publishers such as the New York Times, and usage and sentiment fell when chronological became the default. Your Algorithm lets users see and adjust inferred interests, since LLMs can now describe embeddings in words.
Meta's product teams have shrunk into pods. The canonical team was about a baker's dozen across Android, iOS, server, product, design, data science and research. Pods now have four to six more generalist engineers plus one product staff role blending PM, design, data science and research work.
Engineers now plan and review more than write. Engineering once meant 40 to 60 percent of time writing code, while at labs most time now goes to planning and reviewing. Mosseri has started programming again and finds the tools remarkably good at some things and bad at others.
Token costs may force engineering budget caps. Meta engineers currently have no token limits, and Mosseri calls the token-spend leaderboard a terrible idea. He predicts caps may be needed as a strong engineer's token cost approaches their salary, before a pricing war between frontier models lowers prices.
AI is weak at strategy without hard steering. Mosseri defines vision as the state you want to reach and strategy as an opinionated path there. Good strategy must be controversial, which AI struggles with. Rachitsky argued market data should help, but Mosseri urges caution about outsourcing work to AI.
Product bets that shaped Mosseri's views
He redesigned Facebook's News Feed as an entry-level designer, and it launched in March 2009 to heavy backlash.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
Facebook Home, an Android fork built with HTC, failed.
The first Reels was built on Stories in 2019 and launched in summer 2020, just as TikTok took off during the pandemic.
A video-feed test reached 4 percent of iOS users and drew backlash. He would rather take a backlash every couple of years than stop evolving the product.