Why Your AI-Improved Code Feels Wrong

Overview

This video introduces the concept of intent erosion — the gradual loss of human purpose and feeling from software as AI-driven optimizations accumulate over time. Each individual change is reasonable and measurably improves the code, but the cumulative effect strips away the emotional and experiential qualities that made the product feel intentional.

Key Concepts

What Is Intent?

Intent is not a feature, a requirement, or a spec. It's the why behind decisions — the difference between a system that keeps your notes and one that keeps them the way you want them kept. Users feel intent even when they can't name it. It's encoded in trade-offs, in how code was written one way instead of another, because someone understood the purpose behind it.

The Architect Analogy

Hiring an architect isn't about saying "build me a house." You describe how you want to live in it — your mornings, the light, whether you entertain or want solitude. That's intent, and the architect builds around it. Software should work the same way.

The Ship of Theseus for Software

Like the ancient thought experiment where every board of a ship is gradually replaced, software undergoes constant small changes. As AI-driven iterations accelerate, the original intent embedded in how code was written gets "sanded off" one optimization at a time — not as bugs, but as improvements.

The Western Wear Example

A brand selling cowboy boots builds a customer service agent with a southern accent to carry warmth and identity. Later, someone requests "make it more professional." The AI removes the accent — the fastest path to professional. The brand voice is gone, not because the change was wrong, but because the system didn't know the accent was intentional.

Death by a Thousand Optimizations

No single change breaks anything. No single edit is wrong. But across hundreds or thousands of changes, the human intentions poured into the product are sanded off. The software gets better functionally but flatter emotionally, and you can't point to the moment it changed.

The Human Differentiator

Our differentiator is no longer the ability to write code — that capability is being supplanted. What remains uniquely human is the ability to say what something should feel like, what impact it should have, and why it matters. That's intent. And right now, nothing in our systems protects it.

Three Habits to Protect Intent

Habit 1: Encode Your Intent Into the Product

• Write down the why behind decisions — not requirements, but purpose
• Include: architecture reasoning, desired emotions, brand voice, user experience goals
• Store it inside the codebase, not in a wiki or chat thread that will disappear
• Capture intent immediately — the moment after you make a decision is when you know the why
• If you wait, you'll only be able to test whether something is still green, not whether it's still right

Habit 2: Put Intent in the Path of Your Builders

• Writing intent down isn't enough — the thing doing the building must "trip over it"
• Whether it's an AI or a new team member, they need to see the intent before they start changing things
• Make it impossible to miss: "If you're touching the agent's personality, read this"
• You won't be making every change, and you won't know most changes are being made

Habit 3: Lead with Intent, Not Requirements

• Stop describing features. Start describing purposes.
• Don't say "add notifications" — say why you need them, what problem they solve, what experience you want
• The why guides decisions at the margins — those hundreds of small choices where builders must make assumptions
• With only a feature spec, builders assume functionality. With intent, they assume with purpose.
• This applies to communicating with both AI systems and humans

Key Takeaway

Intent is the soul of what you're building. The people who get good at expressing intent — describing what something should feel like, what impact it should have, why it matters — are the ones who will shape what gets built next. Start encoding your why into your products where it can be found, point your builders at it, and lead with intent every time you ask for something.