Video summary
Head of Instagram Reveals How the Top 1% Actually Work With AI
Main summary
Key takeaways
Summary
Adam Mosseri describes AI as capable but generic “out of the box.” Its usefulness improves when people build feedback loops that let it learn from corrections and outcomes. Because models and tools change quickly, he says using AI well requires continuous experimentation and a learner mindset—not a one-time setup.
AI tools and workflows
- Work command center: Mosseri built a central dashboard using an internal interface, company APIs, a web app, and a simple SQL database. Nightly automations gather information from meetings, email, and Slack, then update tasks, briefings, and meeting history. He says the system helps him see priorities and manage a large workload, though he cautions that most people could use a simpler setup.
- Roger, for follow-ups: This tool tracks requests and promises made in meetings, email, or Slack, and flags items that have gone unanswered or overdue. It helps Mosseri follow up and maintain accountability.
- Email-writing feedback loop: An AI drafts replies but leaves them for Mosseri to review and send. At the end of each day, it compares its drafts with the emails he actually sent and learns from the differences, gradually adapting to his tone, values, and schedule.
- Meeting briefings: An agent summarizes long documents and highlights points Mosseri should know before a meeting. Early briefings were unhelpful; he improved them by having the system review post-meeting materials, such as notes and follow-up messages, and assess whether its earlier briefing emphasized the right issues.
- Shared AI memory: A system called Custodian stores reusable memories and skills and connects to different AI tools. It uses searchable memory files, including concept-based search, so Mosseri can switch tools without repeatedly teaching each one about his preferences and context.
- Personalized podcast app for his children: The app creates audio programs based on the subject, listener age, and desired length. It can use sports-score data, adjust its content as the children age, and offer settings such as voice, model, and playback speed. Podcasts can also be downloaded or shared.
- AI for creators: Mosseri uses AI to surface notable messages and questions in his inbox. He sees potential for tools such as Muse to help creators understand their audience, evaluate content strategy, and spot relevant trends, while noting that some uses still need improvement.
Management, hiring, and measuring productivity
Mosseri uses AI to stress-test decisions, but his broader management approach is to delegate decisions to people closer to the work and make ownership clear. He says this helps capable employees take responsibility and prevents a leader from becoming a bottleneck.
For hiring, he looks for persistence or drive, the ability to learn quickly, and self-awareness, including openness to feedback. He emphasizes learning ability as roles blur and technology changes. He also favors evaluating people and markets by their trajectory, not just their current level.
He cautions against treating AI usage—such as tokens, lines of code, or experiment counts—as a goal. These can be leading indicators, but the important question is whether the work produces real impact and makes Instagram or its products more useful.
Instagram metrics, content, and product features
- Metrics are proxies, not goals: Instagram considers sentiment measures, including users’ satisfaction and whether they feel the platform cares about them, alongside activity such as daily use, messages, Stories, and impressions. Mosseri warns that optimizing too narrowly for one measurable signal can produce poor decisions.
- Views do not tell the whole story: Short interactions are easier to measure than whether content was worthwhile or memorable. For creators, a smaller but more engaged audience may be more valuable than a larger one, depending on their purpose and business.
- Different signals for different audiences: Mosseri says shares per impression are important for reaching people who do not follow an account. For existing followers, watch time and like rate matter more. He also describes content that may re-engage an existing audience without attracting new viewers as potentially valuable.
- Gathering audience insight: He looks for useful feedback in surveys, research, direct messages, comments, and in-person experiences with users in different countries. He describes this as an ongoing process of separating meaningful signals from the volume of feedback.
- Features: Blend creates a shared video feed in a direct-message thread, offering a way to see what content another person may be interested in. Instagram’s interest controls show topics the system thinks a user likes and let the user adjust them.
- Trial Reels: Mosseri says the feature is meant to let creators try a video and gauge its performance before adding it permanently to their profile—not to serve as a formal A/B-testing platform. He acknowledges that creators may want to compare variants, but warns against repeatedly posting nearly identical videos to exploit distribution.
- Prioritization and taste: Mosseri favors doing fewer things well over spreading teams too thin, and prioritizing work connected to Instagram’s purpose and important audiences. He also describes creative choices, such as selecting music for a Story, as stronger when they are intentional and connected to the content.
Guide and product mentions
- AI-agent guide: The host promotes a free newsletter guide intended to help viewers choose a task to automate and set up a first AI agent in about 20 minutes, without coding.
- HubSpot Smart CRM sponsorship: The host describes a workflow in which a call transcript is added to a client record, with suggested deal updates, next steps, and a personalized follow-up email for review.
Main speakers and sources
- Adam Mosseri — Head of Instagram; discusses his AI workflows, management approach, and views on Instagram’s metrics and features.
- The host of Silicon Valley Girl — Interviewer; shares creator-focused examples and asks about AI use and Instagram.
- HubSpot — Sponsor segment describing Smart CRM features.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.