Former VP Design, Meta (Facebook) Founder, Sundial · Author, Making of a Manager
AI ERA
The Willow Tree
Be Sturdy While Being Flexible
"It's always been management's job to manage change. I just think the rate of change is accelerating."
Management = north star + figuring out which resources get you there
It used to be people. Now it's also models.
Rate of change accelerating → sturdy roots matter more than ever
The Builder Role
Dissolve Role Boundaries — We're All Builders Now
Why Julie isn't hiring PMs
At Sundial, engineers prototype and do analysis. Designers do some engineering. "Product science" people blend data, customer success, and product work. The old boundaries are gone.
"I'd love for us to get to the world where everyone's title is 'Builder'. We need to dissolve the boundaries of these traditional roles."
AI lets one person do 10 jobs — the specialist era is ending
Assembling the "Avengers" = matching the right model/tool to each task, like a manager matches people to problems
Google letting go of middle managers = flattening + IC renaissance
Management is still critical — but the people you manage now include AI agents
The meta-skill
Learning to manage AI well uses the exact same muscles as learning to manage people well: clarity, feedback, knowing what good looks like.
Data × Design
Diagnose with Data. Treat with Design.
"Data is not a tool that's going to tell you what you should build. It tells you what's wrong. Design tells you how to fix it."
Most fast-growing AI companies are NOT using data well — they're running on good instincts and good vibes
Legacy companies had years to build logging, growth teams, data infrastructure — AI startups grow before that's in place
What happens: eventually things stop growing. Then you need data. And you don't have it.
The fix: instrument early, build observability before you need it
AI as Teacher
AI Accelerates Learning — Not Just Doing
Use AI to personalize learning: "Explain like I'm 5. Give me analogies." — each person learns differently
Test your understanding by explaining back: "Does this mean X?" — AI gives immediate critique
Engineers at Sundial learning data analysis in 30 min sessions vs. weeks of courses
Just-in-time learning > curriculum-based — learn exactly what you need, when you need it
The underrated use case
We use AI to go faster. We forget to use AI to get better. Learning acceleration might be the highest-ROI use case nobody's talking about.
Contrarian
Leadership & AI Myths Worth Challenging
✗Management is becoming less importantINSTEAD →✓ Management is more important — now you're managing humans AND AI models. The skill set is identical: clarity, feedback, knowing good work.
✗Fast-growing AI companies are data-drivenINSTEAD →✓ Most are running on instinct and vibes. They grew too fast to instrument. The data reckoning is coming when growth plateaus.
✗Hire a PM to drive productINSTEAD →✓ In a small AI company, PM is a role everyone plays. Hire builders who can diagnose with data, treat with design, and ship independently.
✗Data tells you what to buildINSTEAD →✓ Data tells you what's broken. Intuition + design tells you how to fix it. Conflating diagnosis with treatment leads to bad product decisions.