Meta is shifting from 12-person specialist teams to 6-7 person "pods." These are led by a "Product Staff"—a PM generalist who also handles design and data tasks—supported by generalist engineers. This structure increases speed by reducing coordination overhead.
The most effective product leaders don't need to be the source of every idea. Instead, their value comes from having the taste and judgment to curate the best ideas, strategies, and people from the entire team, creating an environment where great ideas can bubble up.
The proliferation of synthetic media won't devalue human content; it will make it more sought after. In a world of abundant AI-generated content, users will increasingly seek out authentic creativity and personal points of view, creating a bigger opportunity for human creators.
A purely chronological feed rewards posting frequency above all else. This creates a system where professional publishers, who can produce content at high volume, inevitably drown out posts from individual users and friends, leading to a worse user experience and lower overall satisfaction.
As AI models improve, detecting AI-generated content will become increasingly difficult. A more sustainable long-term strategy may be to focus on verifying and labeling authentic, camera-captured content. This flips the problem from an arms race of detection to a system of verification.
AI tools can automate tasks that were previously blockers for certain employees. People with great ideas who struggled with the mechanical skills of coding or data analysis can now execute on those ideas, potentially transforming them from low to high performers and leveling the playing field.
With AI handling a large percentage of code generation, the core work of an engineer is evolving. The job now involves spending more time planning what code to generate and carefully reviewing the AI's output, a significant shift from the traditional focus on manually writing code.
With technology changing rapidly, the most successful people will be those who are deeply curious and willing to experiment with new tools, even if it means making mistakes. This "put yourself out there" attitude is becoming more valuable than existing mastery of specific, perishable skills.
For years, recommendation progress came from abstract, "illegible" embedding models that correlate items, not from a deep understanding of user interests like "surfing." Only now are LLMs enabling a shift towards semantic understanding by describing these abstract data clusters in plain language.
The disastrous Facebook Home project taught Adam Mosseri a key lesson: sometimes the best thing to do for a speculative idea is to execute it well. This provides a definitive answer on its market fit, allowing the company to end the project and reallocate resources confidently.
Instagram's first Reels version was built on the Stories platform. This was a mistake because the ephemeral nature of Stories was a poor foundation for a content format designed for broad discovery. The error cost them a year, allowing TikTok to explode during the pandemic.
AI can't generate a great strategy in a vacuum. To get a non-obvious result, a human must provide rich constraints beyond market data, including team motivations, regulatory landscape, and brand identity. The process is more like management than simple delegation.