Skip to content

Head of Instagram Reveals How the Top 1% Actually Work With AI | Adam Mosseri

Silicon Valley Girl: AI, Tech and Career Growth

Hosted by Marina Mogilko

43m
Transcript

Original podcast

Where to listen

Summary

Written by a model from the transcript; not yet reviewed.

In this episode, the guest Adam Mosseri, head of Instagram, joins the host Marina Mogilko to discuss how he uses custom AI command centers, multi-agent setups, and loops to run a massive organization and manage his daily workflow. They examine the practical realities of working with artificial intelligence, moving past vanilla prompts to train agents that learn from edits and delta comparisons. The conversation also covers broader strategic themes, including how Instagram measures user sentiment and ranking, why trajectory matters more than static metrics in hiring, and how creators can balance raw view counts with genuine audience impact.

What makes the episode worth listening to is the granular, behind-the-scenes look at actual operational tooling used by a tech executive managing three billion users. Rather than speaking in abstract theories, the guest walks through specific internal systems like Roger for tracking delegated tasks, Custodian for synchronizing memory across different AI models, and automated evening briefings that summarize lengthy documents. It provides actionable frameworks for anyone looking to build a personal AI harness and rethink productivity in a rapidly shifting technological landscape.

  • Build feedback loops that save your drafts and compare them against your final edits so your AI agents can learn your authentic tone over time. 7:50
  • Delegate decisions publicly to capable people close to the work so you avoid becoming a bottleneck for your team's progress. 16:55
  • Prioritize hiring for a combination of grit, hustle, quick learning ability, and self-awareness rather than static current experience. 18:20
  • Focus on second-order preferences and harder-to-measure proxies like watch time and shares per impression rather than shallow metrics like raw views. 24:49

Chapters

  1. 0:00
    Episode Introduction and Host Greeting

    Marina Mogilko opens the episode, setting the stage for a deep dive into AI's impact on work and creativity.

  2. 0:57
    Building and Refining AI Loops

    Discussion on how Adam built AI loops, improved briefings, and embraced a learner mindset.

  3. 5:12
    Desk: Unified Command Center

    Explanation of the Desk interface that consolidates tasks, briefings, and AI agents into one view.

  4. 7:53
    Creating AI Agents and Feedback Loops

    Adam details how he trains AI agents like Muse by feeding them drafts and learning from edits.

  5. 9:00
    Hiring Philosophy and Talent Development

    Adam shares his hiring criteria—grit, quick learning, self‑awareness—and how he supports growth.

  6. 20:00
    Future of Instagram and Feature Priorities

    Adam discusses upcoming features, testing practices, and the importance of aligning growth with platform values.

  7. 21:51
    Measuring Impact Beyond Tokens

    Discussion on how Instagram measures sentiment, engagement, and long‑term value instead of token usage.

  8. 26:35
    Content Strategy and Audience Metrics

    Adam explains how he balances view counts, watch time, and emotional impact when evaluating content.

  9. 37:20
    Creator Tools and Muse Features

    The host explores Muse’s capabilities for creators, including audience insights and trend analysis.

  10. 43:52
    UKG Workforce Intelligence Pitch

    A promotional segment highlights UKG’s AI-driven solutions for staffing, pay, and compliance, emphasizing real-time decision making.

  11. 44:47
    Music Selection Philosophy Discussion

    The host probes how Instagram’s content creators choose music, exploring intentionality and cultural relevance in storytelling.

  12. 45:49
    AI and the Learner’s Mindset

    The conversation turns to how AI reshapes continuous learning, raising stakes for staying relevant in tech and beyond.

Highlights

  • 1:06

    Building An AI Command Centerstory

    The guest explains how he built his own AI command center called Desk that processes meetings, emails, and messages every night to automate his workflow.

  • 3:14

    Tracking Accountability With Rogerstrategy

    The guest details an internal tool he calls Roger that tracks commitments made in meetings, emails, and Slack, and prompts him to follow up on unfulfilled requests.

  • 7:17

    Training AI Via Feedback Loopsstrategy

    The guest explains that out-of-the-box models are average, but building loops where agents compare their drafted emails to your final edits allows them to learn your tone.

  • 15:22

    Connecting Models With Custodianstory

    The guest shares how he wires different AI tools like Claude, ChatGPT, and Meta AI into a shared memory system called Custodian so they can benefit from unified context.

  • 17:12

    Delegating Decisions Publiclyinsight

    The guest argues against hoarding decisions, explaining that he publicly assigns choices to capable team members to empower them and prevent himself from becoming a bottleneck.

  • 22:02

    Metrics Versus Actual Goalsinsight

    The guest warns that large companies frequently conflate metrics with goals, noting that metrics are merely proxies for underlying objectives expressed in prose.

  • 32:01

    Trajectories Over Static Pointsinsight

    The guest advises that in careers and hiring, a person or market's trajectory matters much more than where they stand today.

Quotes

  • “If you want to make AI work for you, you have to embrace that learner's mindset.”
    Unnamed speaker · 1:29
  • “I think a huge thing that matters for unconnected reach, so reaching people who don't follow your account, is definitely sends per impression.”
    Unnamed speaker · 38:56
  • “I think AI has increased the value of being willing to try and learn new things significantly”
    Unnamed speaker · 46:07
  • “Like you can learn to learn.”
    Unnamed speaker · 46:36
  • “I feel like it's forced me to experiment a lot more and try a lot more things.”
    Unnamed speaker · 46:40
  • “And I've actually, as a result, gotten quicker at picking up new things.”
    Unnamed speaker · 46:46
  • UKG 43:47–44:17
  • Progressive 44:28–44:43

About this episode

In this episode, I talked to Adam Mosseri, head of Instagram, about the AI tools he has built for his work and his family. He walks me through his command center for meetings, tasks, and follow-ups, and explains how he gets AI to write emails that sound like him. Instead of correcting the same mistakes every time, he has it compare its drafts with the emails he actually sends and learn from the differences. He also shows me the podcast app he built for his kids and the shared memory system that lets him switch AI tools without starting from zero. We talk about the three qualities Adam looks for when hiring, why using more AI doesn’t necessarily mean getting more done, and how he decides what Instagram should build next. He also explains which engagement metrics matter for reaching new audiences, why views don’t tell the whole story, and what Trial Reels were actually designed to do. 📩 Your first AI agent in 20 minutes — get the guide through my free newsletter: https://siliconvalleygirl.beehiiv.com/p/your-first-autonomous-agent-in-20-min-5429?utm_source=spotify&utm_medium=video&utm_campaign=futureproof-sub&utm_content=first-agent&utm_term=adam-mosseri https://siliconvalleygirl.beehiiv.com/p/your-first-autonomous-agent-in-20-min-5429?utm_source=spotify&utm_medium=video&utm_campaign=futureproof-sub&utm_content=first-agent&utm_term=adam-mosseri 𝕏 : https://x.com/siliconvalleymm 🔗 Instagram: https://www.instagram.com/siliconvalleygirl/ 💼 LinkedIn: https://www.linkedin.com/in/marinamogilko 📌 My Companies & Products: https://Marinamogilko.co

Transcript

0:00
43:21

9,634 words · 4 speakers

Speaker names marked are inferred from speaking time and the show’s known hosts — they are not voice-verified.

Topics

  • AI
  • Instagram
  • Muse
  • email
  • Blend
  • Trial Reels
  • Meta
  • Custodian
  • People First AI
  • Slack

Page updated · transcript and summary are machine-generated