Netflix Engineering • AI Engineer Summit 2024

The Infinite
Software Crisis

"I've shipped code I didn't quite understand. Generated it, tested it, deployed it. Couldn't explain how it worked. I'm willing to bet every one of you have, too."

Jake Nations

Netflix • 00:00:28

Watch this moment

Minutes

Tokens Analyzed

Words Specified

Rich Hickey

Framework

The Crisis

Understanding vs Generation: What we're sacrificing for speed

The Hard Part Was Never The Mechanics

"The hard part wasn't ever the mechanics of coding — the syntax, the typing, the boilerplate. It was about understanding the actual problem and designing the solution."

— Jake Nations (00:03:44)

Watch explanation

Before AI

Friction forced understanding

Manual work revealed constraints

Mistakes taught lessons

With AI

No friction = no understanding

Generation replaces design

We ship what we don't understand

Easy vs Simple

Rich Hickey's framework: Understanding the real trade-off

The Rich Hickey Framework

Easy = Near

Proximity to your capabilities: What you can do right now without learning

• Familiar tools

• Known patterns

• Comfortable workflow

Simple = One Braid

Structural composition: How many concepts are intertwined

• Single responsibility

• Clear separation

• Minimal complexity

"Every time we choose easy, we're choosing speed now, complexity later. AI has destroyed that balance because it's the ultimate easy button. And it makes the easy path so frictionless that we don't even consider the simple one anymore."

— Jake Nations (00:04:56)

Watch the critical insight

AI Eliminates The Resistance

"There's no resistance to bad architectural decisions. The code just morphs to satisfy your latest request."

— Jake Nations (00:06:37)

Watch (00:06:37)

The Complexity Trap

A real example: Netflix's authorization refactor

Netflix Authorization Refactor

Changing how authentication works across services

The Challenge

Netflix needed to change how authorization worked across hundreds of microservices. This wasn't just about updating code — it was about understanding how authentication flowed through their entire system.

Step 1: Research - Analyzed 5 Million Tokens

Used AI to read through their entire codebase and documentation

Step 2: Plan - Created 2,000 Word Specification

Synthesized findings into a comprehensive design document

Step 3: Validate - Manual Migration Revealed Constraints

Did the work manually to find what AI analysis couldn't see

Tokens Analyzed

Words Specified

"Five million tokens became 2,000 words of specification."

— Jake Nations (00:10:47)

Watch (00:10:47)

The Solution

Three phases to maintain understanding while leveraging AI

Phase 1

Research

Let AI explore the codebase. Read documentation, analyze patterns, identify relationships.

Tool: Use AI to analyze what would take weeks manually

Phase 2

Plan

Synthesize research into a specification. Design the solution before coding.

Output: 2,000 words of clear specification

Phase 3

Validate

Do the work manually. Find hidden constraints. Validate the plan against reality.

Goal: Highest leverage moment — catch errors here

The Highest Leverage Moment

"This is where you validate the analysis against reality. The highest leverage moment in the entire process. Catch errors here. Prevent disasters later."

— Jake Nations (00:12:14)

Watch why validation matters

Why Manual Migration Was Crucial

"That manual migration was, I'll be honest, it was a pain, but it was crucial. It revealed all the hidden constraints, which invariants had to hold true, and which services would break if the auth changed. Things no amount of code analysis would have surfaced for us."

— Jake Nations (00:14:57)

Watch (00:14:57)

Human Agency

What only humans can contribute

Pattern Recognition

AI can generate code. It cannot recognize when you're solving the wrong problem. That instinct — "Hey, this is getting complex" — is uniquely human.

"That instinct that says, 'Hey, this is getting complex.' It atrophies when you don't understand your own system."

— Jake Nations (00:17:05)

Problem Framing

Before writing code, someone must frame the problem. AI is a tool for solving problems, not identifying which problems matter.

"The developers who thrive won't just be the ones who generate the most code, but they'll be the ones who understand what they're building."

— Jake Nations (00:18:11)

"The developers who thrive won't just be the ones who generate the most code, but they'll be the ones who understand what they're building, who can still see the seams, who can recognize that they're solving the wrong problem."

— Jake Nations (00:18:11)

Watch (00:18:11)