Managing people and managing AI require the same core skills

Executive overview

The skills that make a great manager — defining clear outcomes, assembling the right resources, and running good process — map directly onto working with AI agents. As AI flattens org structures and empowers individuals to do more, every knowledge worker is becoming a manager of sorts.

Julie Zhuo argues that the future belongs to builders: people who dissolve traditional role boundaries and use AI to cover the gaps. The same human fundamentals — clarity of goals, candid feedback, conviction in purpose, and a win-win mindset — determine whether teams and AI systems deliver great work.

Managing AI is managing people

• The three pillars of management — people, purpose, and process — apply unchanged to agentic systems.
• The hardest part of prompting AI well is the same as managing humans: getting crystal clear on what success looks like.
• Writing evals for AI mirrors the discipline of setting objective KPIs; both force you to define done.
• Different models have different strengths; assembling the right combination is an Avengers problem.
• Treating agents with care may matter more as systems grow more capable — the habit costs nothing.

The builder model: dissolving traditional roles

• AI lifts most individuals from the 0th to the 60th–70th percentile in unfamiliar skills, making specialised hiring less necessary.
• Sundial eliminated the PM role; removing it forced engineers to own product definition rather than delegate it.
• Front-end/back-end distinctions are also blurring: engineers take on cross-discipline work with AI support.
• Initial productivity dips when roles expand, but the compounding return is a more versatile team.
• The key hiring question shifts from "what is this person's role?" to "what specific skills does this project need?"

Using data well: diagnose with data, treat with design

• Most fast-growing AI companies are running on instincts and good vibes — they lack the logging infrastructure to do real analysis.
• Data's job is to reflect reality back, not to tell you what to build; that remains a creative act.
• Conversational analytics require new methodologies: click-based metrics don't capture intent in LLM-driven products.
• Quantitative metrics carry false precision — choosing which metrics to track and interpreting a 5% move are both judgment calls.
• Every great designer is obsessed with understanding reality; data is just the most systematic way to do that.

Managing change: the willow tree

• The rate of change is accelerating faster than any prior decade; managing change has always been a manager's job, but the speed is new.
• Fear and opportunity are two sides of the same coin — leaders who frame change as exciting will outlast those who find it threatening.
• The willow tree metaphor: sturdy enough to survive storms, flexible enough not to snap.
• Pretending uncertainty doesn't exist destroys trust; naming it — "yes, it's hard, we will work through it" — builds it.

Managing yourself first

• Every strength is its own weakness; every weakness signals a corresponding strength.
• Think in infinite dimensions: no single dimension is your identity or your worth.
• Knowing your natural wiring lets you read context — when to lean into decisiveness, when to lean into thoughtfulness.
• ICs should deepen craft; managers need breadth. Which path to take depends entirely on your actual goals.
• Suffering usually means your goals and your daily actions are out of sync.

Feedback as a daily practice

• Teams that get 1% better per week will dramatically outperform teams that get 1% better per month.
• Feedback is calibration: it reflects back what you can't see about yourself, closing the gap between self-perception and reality.
• Establish the feedback relationship before anything goes wrong — get explicit opt-in from day one.
• Before giving hard feedback, check your own intent: if the goal is to be right rather than to help, it will not land.
• Naming your nervousness out loud — "I'm worried this will affect our relationship" — does more work than finding the perfect phrasing.

Purpose and conviction

• Middle managers who don't personally believe in a strategy will not execute it well — there are no exceptions.
• When you disagree with a directive, engage in dialogue to decompose it into specific hypotheses; you will likely find something you believe in.
• Disagree and commit becomes much easier when you have isolated the one assumption you're unsure about and designed a small test to validate it.
• Sharing uncertainty with your team ("I don't know for sure, but here's why we're testing it") is more effective than false conviction.

Win-win as a management philosophy

• If getting better outcomes requires someone else to lose, the strategy is already broken.
• Letting someone go can be a win-win: prolonging a bad fit is worse for the individual than ending it honestly.
• The goal of a performance conversation is not to pressure someone into compliance; it is to find alignment between their goals and the team's.
• Framing a let-go as "this isn't the place where you'll do the work you're most proud of" is not spin — it is usually accurate.

Navigating the AI era as a human

• Emotional regulation is the most important skill to build for an AI-native world.
• AI's greatest risk is comfort: it can make distraction, validation, and avoidance infinitely frictionless.
• True freedom is choosing your own challenges, not eliminating challenge entirely.
• The hardware of human biology — the need for difficulty, pride, and meaning — is not changing with the tools.