> For the complete documentation index, see [llms.txt](https://jamiewen00.gitbook.io/ai-engineering-handbook/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://jamiewen00.gitbook.io/ai-engineering-handbook/building-ai-products/fuzzy-land.md).

# Fuzzy Land

> Inspired by [Rethinking how we Scaffold AI Agents](https://www.youtube.com/watch?v=-rsTkYgnNzM\&t=523s)

Turning apps into AI systems creates fuzzy lands. When should you add fuzziness to your app?

## Fuzzy Zero

All exact land. No fuzz. Every step is set. User input is strict. Output never changes.

<figure><img src="/files/jE464RoYEJw21zFrwiOz" alt="image"><figcaption></figcaption></figure>

## Fuzzy UI

A small UI part is fuzzy land. Users type free text. Backend stays strict.

* Pros: UI feels smart.
* Cons: User input is wild. Output is same old, so what's the point? 🤷‍♀️

<figure><img src="/files/etXJ3gnfxRTMGp1g7XfX" alt="image"><figcaption></figcaption></figure>

## Fuzzy UI: Clarify

Add steps to clarify what users want. Bridge fuzzy UI and strict backend. Set clear rules up front.

* Pros: Less confusion. Clear rules.
* Cons: More steps for users.

<figure><img src="/files/2wsFlPypgvNHY88zD6iG" alt="image"><figcaption></figcaption></figure>

## Fuzzy API

Add a small fuzzy land to the API. For example, click to get insights.

* Pros: Fuzz is small. Most of the app is clear. Users know what to expect.
* Cons: Output is safe. Less wow.

<figure><img src="/files/4SC2bKky9mXMOBZRBE3Z" alt="image"><figcaption></figcaption></figure>

## Constrained Flow

Exact-led fuzziness.

* Pros: Reliable. Good for the right job.
* Cons: Feels like workflow or automation, not very agentic. (if it does the job, why bother?)

<figure><img src="/files/l2PXtL2hlw6CglZNaNHl" alt=""><figcaption></figcaption></figure>

## Fuzzy Flow

Make the whole app a fuzzy land. It may contain some exact lands but it's a fuzzy-led experience. AI runs it all.

* Pros: If it works, it's huge. My attempts stay as fun prototypes.
* Cons: Hard to steer. Costly. Wild.

<figure><img src="/files/1VazHAXuCjEFiPndAxPW" alt="image"><figcaption></figcaption></figure>

## Appendix

My main takeaways from [Rethinking how we Scaffold AI Agents](https://www.youtube.com/watch?v=-rsTkYgnNzM\&t=523s):

<figure><img src="/files/r1OkkYFRklCgIAnXGwwm" alt="image"><figcaption></figcaption></figure>

<figure><img src="/files/w7h8q0KMWgERA7uwHwhR" alt="image"><figcaption></figcaption></figure>