Designing AI Solutions for People

Dennis Hölzenbein: It is often used for chat or voice interfaces, such as AI agents. There are also more and more software tools that incorporate AI, for example for translation or for generating images and videos. AI does not necessarily have to be visible. It can also be integrated into applications in the background, for example for automatic text suggestions in e-mails or for automating processes. 

Dennis: In the past, chatbots could, to put it simply, answer questions and perform a few tasks. But as soon as a request became more complex, chatbots no longer understood the user's intention and quickly reached their limits. With good, but also extensive design, it was possible to achieve a reasonably good user experience, but these were the exceptions, most chatbots provided poor experiences for the users who had high expectations and were disappointed. 

That has now changed. AI agents understand users much better – with the help of their LLM's training data and the additional context they are given. The context comes from the data provided, for example from documents and websites, and of course from user prompts. In addition, system prompts define the role, personality, and basic behavior of AI agents. Depending on the desired level of autonomy, AI agents can then plan and execute several steps independently. They don't just react, they work proactively: they suggest solutions, take on tasks, and deliver results that have value. Another important difference from earlier chatbots is that today, these agents can be designed and implemented more efficiently using low-code or no-code tools.

Dennis: First, AI should be used where it is truly helpful. Therefore, identifying the right use cases is an important foundation. AI must integrate well into existing processes – or be the catalyst for rethinking poor processes. The value of using AI should be quickly apparent to users. They need to feel as directly as possible that AI supports them, but does not patronize them. Another aspect is that people want to remain in control. That's why the user guidance in the interface must be designed in such a way that people can understand the actions of the AI and intervene if necessary. Verifiable sources on which the AI's results are based are also important. 

All of this builds trust. And to ensure that trust is not immediately lost, good AI onboarding is important: make it clear from the start what the AI can do, where its limits lie, and what example prompts look like. This is because the first hurdle arises as soon as users are faced with a chat input field, and don't yet know what the AI is capable of and what they should enter. This important AI onboarding is an example of various design patterns that help to design products in such a way that people trust AI more and want to use it more.

Dennis: AI is often used in agents or workflows where humans are still involved, making decisions and ensuring quality. For example, in content creation. Many steps, from brainstorming to creating content with images and text for various touchpoints, can be significantly accelerated or partially automated by AI. 

All of this works when standards such as brand, design, and language guidelines, including best practice examples, are available. Ideally, this data should be as easy to interpret as possible. This is because large language models can only process information as well as it has been prepared. If documents are written in understandable, user-oriented language and structured in a meaningful way – for example, with clear headings, paragraphs, and semantic markings – it is easier for AI to clearly recognize content and deliver high-quality results. 

AI agents are also increasingly being used in customer service or to access internal company knowledge. This makes expertise available more quickly and reduces frustration – for example, when an AI agent answers support requests or shows employees the appropriate internal policy directly.  
 
But the possibilities go far beyond that: AI agents can do more than just provide information. They access tools that allow them to handle more complex use cases. Just as an example: an agent can research information from various sources, store results in databases, combine them with knowledge from the company, and generate content that is displayed in a defined format, such as a report or presentation. Another feature of AI agents is the ability to remember, learn, and provide individualized support to users in their tasks.

Dennis: Selecting the right use cases that bring real benefits to people and companies. And ensuring that the technology or AI engineering reliably and repeatedly delivers what AI promises. Therefore, it seems most promising to first try smaller improvements and more limited use cases before attempting, for example, to fully automate complex processes or diverse tasks via AI or design AI agents that act fully autonomously. As with the design of all products, and even more so with AI, it is important that we design for people and their needs so that technology leads to positive change and helps people advance in what they do. 

Dennis: We support organizations in many phases of AI product design. This begins with workshops in which we gather the perspectives of a wide range of stakeholders, select use cases, visualize and evaluate ideas. We also conduct qualitative interviews to understand people's needs and attitudes toward AI, as well as their existing processes. This forms the basis for truly human-centered product development. 

Building on this, we use prototypes to gather early feedback from users – for example, on use cases and initial ideas, or even on specific requirements for the interface of an AI product. This ensures that solutions are relevant and intuitive to use. 

In UX creation and UI design, we design based on AI design patterns. For example, we remove entry barriers, build trust, and design solutions so that people and AI can work well together. In addition, we develop the personality, tone of voice, and behavior of the AI. 

In the content area, we advise on AI-supported editorial processes, structure information, and create user-oriented texts. This is precisely the basis for ensuring that AI always has the latest data available, which it can interpret optimally. 

In one sentence: We design so that people and companies can make the best possible use of the opportunities offered by AI.