Introduction

Did you know that in one of the most advanced AI agents today, only 1.6% of the code is “pure intelligence”?

The remaining 98.4% is not AI. It’s infrastructure: context management, permissions, safety layers, and recovery logic.

I recently read the paper “Dive into Claude Code: The Design Space of Today’s and Future AI Agent Systems” (Liu et al., 2026), and there’s one idea that feels especially relevant for design:

The real product is not the model. It’s everything around it.

When AI stops being the center

For years, we’ve designed products where logic was relatively predictable.

Not anymore.

With agents, systems:

• decide
• interpret
• execute
• fail

And most importantly, they do so in non-deterministic ways.

That completely changes where design lives.

1. Design is the guardrail, not the prompt

One of the most interesting points in the paper is that architecture is driven by human values:

• user control
• safety
• reliability
• adaptability

This is not just engineering. This is design.

The challenge is no longer defining perfect flows. It’s designing systems that know what to do when things go wrong.

Because they will.

We design fewer “paths” and more “recovery”.

2. Infrastructure is the experience

If 98% of the system is rules, permissions, and logic, then experience depends on that.

Not on the UI.

For example:

• when does the system ask for permission?
• how does it explain what it’s doing?
• what happens when it fails?

These decisions don’t live in Figma.

But they fully define the experience.

3. Designing systems, not screens

This is the biggest shift.

If you don’t understand:

• how context is managed
• how tools are routed
• how recovery works

you can’t properly design the product.

Not because you need to code, but because you’re designing something non-linear.

And that changes how we work:

• closer collaboration with engineering
• more system-level decisions
• less obsession with screens

So where is the value?

This is where the 1.6% becomes interesting.

If “AI” is only a small part of the system…

👉 the value is not in the model 👉 it’s in how you control it

🚀 My conclusion

AI is the engine.

But the product is:

• the chassis
• the brakes
• the steering

A powerful engine without control isn’t innovation. It’s an accident waiting to happen.

Closing

The conversation around AI in design is heavily focused on tools and models.

But maybe the more important question is:

Who is designing the system that controls all of this?

Because that’s where the experience is truly defined.

And where we can create the most value as designers.