Making generative AI concepts approachable

How UX is evolving for AI developers

User experience (UX) for AI developers is evolving with the advent of generative AI. Why? Until recently, developing AI applications in the enterprise was a niche specialization most often performed by people with PhDs in machine learning. As such, UX catered to a specialized audience of machine learning engineers and data scientists, who brought with them a shared technical understanding and language of machine learning. However, gen AI has changed the audience that we’re building for: We’re now seeing full stack software developers, technical leads, business users, and others trying and exploring AI products. And this new wave of customers brings their own understanding and language around AI.

At Google, through collaboration across UX design, writing, and research, we’ve been working on ways to evolve our UX discipline to embrace emerging AI users on our cloud platform. Through user research programs and outreach to hundreds of AI developers from every corner of the enterprise world, we’re gaining a deep understanding of how to evolve our UX standards.

If you’re developing or serving applications with gen AI or asking employees to interact with gen AI, you should also consider how to evolve your UX practices so that your gen AI implementations are successful. In this blog post, we’ll share some UX principles that we’ve developed at Google to make gen AI more accessible to our new audience and how we’re putting them into practice.

1. Learn by doing

In our gen AI user research programs, we’ve heard from developers that they want to experience the value of gen AI immediately. In other words, they want the experience put up front, without having to read a lot of documentation, understand technical jargon, or set up complex underlying infrastructure.

With gen AI, therefore, we encourage users to learn through action by building experiences that focus on play and experimentation. In doing so, users can absorb a lot simply by playing around with settings and levers in a safe environment that won’t break something in production. For example, temperature — a hyperparameter that controls the degree of randomness of a model's output — has a technical explanation, but users tend to grasp its effect more readily when they can see the minimum and maximum effects of the setting in action.

One place we’re putting the “learn by doing” principle into practice is in Vertex AI’s Model Garden. Model Garden offers hundreds of gen AI and traditional machine learning models, including our family of Gemini models (like Gemini Flash and Gemma), open-source models like Llama from Meta, and partner models like Claude from Anthropic.

To make the model selection experience more welcoming and help customers start building with AI as quickly as possible, we have embedded experiences in our model cards that are similar to sandbox testing environments.

These “playground” experiences allow customers to directly interact with a model, tweak parameters, and quickly determine whether the model is suitable for their use case. They can also get a feel for models using prompts that are important to their business before launching code in a Colab Enterprise notebook or environment of choice.

2. Use plain language

A frequent customer feedback theme that comes up in our user studies is the accessibility of AI, particularly technical jargon. Traditional machine learning jargon is daunting and often a barrier for new audiences embarking on a journey with AI for the first time. And while jargon can be useful for concisely communicating complex ideas, it only works if your audience has taken the time to learn and retain that information. For newcomers, jargon is exclusionary and dissuades them from engaging further with your products.

As UX practitioners, it’s important to design with plain language so that all users can access gen AI capabilities and understand how they work. Jargon can exist, but it should be introduced through the learning experience. Our user research has consistently shown us that most users don’t want to read documentation before they try something; similarly, users don’t want to demystify a bunch of terminology before they jump into an experience.

To help users engage with gen AI without technical jargons, we’ve created interactive learning pathways that use real resources so users really can learn in a controlled environment that slowly introduces technical concepts. We also ensure these short, in-context tutorials let you bring along your own AI use case, so you’re not just trying another “Hello World” tutorial that has no context to your own interests. For example, the “Tune a foundation model” tutorial provides you with a tuned model based on your own dataset within 20 minutes, during which you learn key concepts and get familiar with Vertex AI.

These tutorials focus on starting with simple language and eventually connecting them to their related technical terms, enabling users to strengthen their mental model while working. In our internal studies, we found this approach delivered measurable improvements in speed, task completion, and user satisfaction for complex tasks that would otherwise require weeks or months of education.

3. Get to code quickly

The more time we spend designing for emerging AI developers, the more we know how important it is for them to quickly jump into code. This also aligns with our “learn by doing” approach: users want to try things out, learn how they work, and then quickly jump to their preferred development environment.

To support this principle, we added a one-click “Get Code” experience in Vertex AI Studio’s prompt editor. This experience allows developers to take the prompt working on — including the gen AI model’s settings and parameters — and transform it into an SDK or API request in Curl, Python and other programming languages. You can even deploy it to a Colab Enterprise notebook with one click, allowing developers to jump from experimenting in a user interface to developing with code in just one click.

Putting gen AI to work

We’re committed to evolving the UX of AI to empower every enterprise developer. Across our entire product portfolio, we’re applying the foundational principles of encouraging learning through doing, using simple language, and getting to code quickly to make this ground-breaking technology accessible to everyone. We’re excited to see what Google Cloud customers can build with our gen AI products, so we’re working hard to make them as engaging, easy-to-use, and valuable as possible from the moment you start using them. To learn more, visit Vertex AI Studio on Google Cloud.