Product Design

AI shouldn't feel like magic

Beyond the charm: the pitfalls of 'magic buttons' and how Zylon brings clarity to AI for everyday use.

BY Zylon - 
January 2024

Magic is fun

Everybody loves a good magic trick. Witnessing something we don’t understand and can’t control makes us feel fascinated. We suspend our disbelief for some minutes and accept everything and anything is possible.

That very nature of magic is what makes AI demos so entertaining and appealing. When an unexpected result appears on screen and we can’t explain how it is even possible that it works, the wow effect is guaranteed.

But the same reason that makes magic so fun and attractive is what makes it a very bad user experience choice for day-to-day products.

The magic button anti-pattern

As existing products rushed to include AI-features in their features portfolio, most took a product-design shortcut and, instead of understanding the problem being solved -more on this later-, they jumped right into the solution. The result was most of the times the addition of a brand new button to their user interface, featuring a magic wand 🪄 or magic sparkles ✨, claiming user's attention.

As a user, clicking on that new magic button feels exciting at first, but rapidly becomes disappointing after a couple of tries. Just like attending a magic show, you don’t have control of what will come out of the hat each time the magic happens, with the difference that in this case you are trying to get some real work done.

The main reason why a magic button is a user experience (UX) anti-pattern is that it goes against a key concept in UX: user expectations management. As expressed in this great article, expectations management is the process of:

1. Understanding what users have in mind as a successful outcome;

A magic button extracts very little if any information about the user's intention. The user doesn’t know what to expect and the product tries to guess what the user may want.

2. Matching it with reality (is it feasible?);

Given the input is so undefined and broad, the odds of the result being incorrect, incomplete or plainly wrong compared to the user’s intention are too high. It is no surprise that most products using this approach include a message like “AI responses can be inaccurate or misleading” in the results screen.

3. And guiding them in the process, so that the end result is as close to what they wish for as possible. And even if something is missing, they know why and are okay with it;

The magic button interaction is a one-shot interaction: click, cross fingers and get a result. There is no refinement or informed process, generating a big misalignment between the user intention and the end result.

The result is most of the times a low perceived value which impacts user retention.

A chat is not always the answer

While generative AI enables natural language bi-directional interactions, exposing a chat UX where the user asks for the required task through a textual query shares some of the same issues that the magic button introduces.

Although users should have the ability to express their intentions more precisely through open-text input, the reality is that most of the time, they are uncertain about how to do so. This is why prompt engineering becomes necessary to obtain accurate results from ChatGPT-like applications, acting as a barrier to adoption.

The result returned by the chat can be more or less accurate based on the prompt, but it is still a one-shot interaction, forcing users to understand how to follow up with subsequent interactions to refine the result.

That’s why a chat UX is great for certain use cases such as customer support or simple question & answers widgets, but falls short for more complex interactions.

Understanding the Job to be done

The first step of every good product design is understanding the problem being solved or, following a popular framework, the user’s Job to be done. Without understanding the user's need we are trying to cover, it is impossible to offer a good solution.

As product development teams, it is key that we acknowledge that, despite the hype, AI is not an end-goal by itself, but just a -powerful- new tool we can use to solve user’s problems.

This is a topic that goes beyond the scope of this article; we’ll be covering how to apply product design best-practices to AI product development in future posts.

Zylon’s “magic-less” approach to product development

As explained above, while magic is fun and entertaining, it is not the type of user experience suitable for everyday products.

Zylon offers a product designed to fix the usability problems introduced previously. It abstracts the complexities of AI technology, and proposes a fresh approach to user experience where available actions map to real use cases (query company data, summarize, generate a structured report) and are simple, consistent, transparent and reproducible.

That design pattern is applied to both user-initiated interactions and proactive interactions in which Zylon autonomously executes tasks to help users achieve their goals - we’ll share more about AI proactivity in future posts too.

We are proud to be building a product that doesn’t feel magical, even if that implies achieving less of a wow initial reaction. We pursue a different wow effect, the one that appears when our users look back at the amount of time and resources saved by using Zylon to optimize their every day tasks and processes.

Want to experience it yourself? Join the beta today!

Iván Martínez, Co-founder & CEO

Product Design
BY Zylon - 
January 2024

Get Started Today

Join the Beta