ChatGPT: Redefining Design Beyond AI Integration

The Emergence of ChatGPT and the New Era of Design

The arrival of ChatGPT has marked a decisive turning point in the way we interact with digital technologies. For many, it has been a revelation, an awareness that we are entering a new era where machines no longer merely follow precise instructions. Instead, they can understand our intentions and act accordingly. This paradigm shift has been brilliantly described by Jakob Nielsen in his article on AI as the first new user interface paradigm in sixty years. For decades, we dictated to computers what they should do, command by command. Today, AI allows us to simply express our wishes, and it takes care of figuring out how to fulfill them. This is not just a new feature, but a fundamental transformation of the very nature of computers.

The Rise of AI Assistants in Products

Following the introduction of ChatGPT, companies rushed to integrate AI assistants into their products. Whether in enterprise tools, creative software, or cloud platforms, chatbots and copilots have become ubiquitous. Three years later, these assistants have proven their utility by enabling users to ask questions rather than navigate through complex interfaces, obtain instant answers, and delegate repetitive tasks. This has brought undeniable added value.

However, one problem persists: in most cases, if the AI assistant is removed, the product remains functional. This shows that AI is often merely overlaid, rather than truly integrated into the core of the product. Adrian Levy highlighted this point with his concept of the integrated intelligence test, which the majority of current products fail.

The Limits of Augmented AI

The analogy of the Incredible Hulk opening a jar of pickles aptly illustrates the current situation: while AI brings real power, it is not being utilized to its full potential. To understand what is lacking, it is essential to examine how users learn to use products. Each product imposes a learning threshold, which is the minimum knowledge required to start using it.

Take a door, for example: at some point, we all learned that we need to turn the handle to open it. In the digital realm, this threshold is divided into two layers: the conceptual layer, which concerns the fundamental ideas of the product, and the interaction layer, which pertains to how these ideas are represented in the interface. Both layers must be crossed to use the product.

The Challenges of Enterprise Software

For consumer software, the conceptual threshold is often thin, as the ideas are generally familiar to users. However, enterprise software presents a different challenge. Designers of these products often create complex concepts to structure functionalities and make them tangible. Before the era of AI, this was necessary, even if it introduced a high learning threshold. Once this threshold was crossed, these concepts genuinely helped users navigate the product.

Take AWS as an example. To store data in the cloud, it is imperative to understand concepts such as accounts, services, and resources. These concepts do not exist in the real world but were invented to organize AWS's offerings. They represent a toll that users must pay to access the product.

The Limited Impact of Current AI

Adding an AI assistant to AWS without modifying the underlying product can enhance the user experience. For instance, the assistant can explain concepts or execute tasks on command. However, this does not free users from the need to learn AWS's specific language. The conceptual burden remains on their shoulders.

This illustrates the limit of augmented AI: it improves efficiency within the existing conceptual framework but does not transform it.

Towards a Truly AI-Native Experience

Imagine a user wanting to store data for an application. In a truly AI-native product, the user could simply express their need in natural language, and the AI would act as a translator between this intention and the product's technical concepts. The AI would ask questions to clarify the intention, map the goal to the appropriate concepts, and configure the system accordingly. The user would never need to understand concepts such as services or resources.

Reducing the Conceptual Burden

Of course, this does not entirely eliminate the learning threshold. The user still needs to grasp basic concepts, like cloud storage and its utility. However, AI can reduce the additional conceptual layer that exists solely because of how the product was designed. Concepts like accounts and services only make sense in the context of the product and can be managed by AI.

AI Beyond Initial Configuration

AI can continue to carry this conceptual burden throughout the product's lifecycle. Once storage is configured, the user will need to manage it. In a conventional product, this means learning new vocabulary. For example, an error message might state: "Your S3 bucket has exceeded its storage quota. Check your IAM policy to update access permissions." The user is then confronted with AWS's technical language.

In an AI-native product, the message would be simplified: "Your storage is almost full. Here’s how to increase it." The AI continues to translate technical concepts into accessible language.

Trust, an Essential Pillar

It is crucial to clarify that the goal is not to replace traditional interfaces with AI chats. Classic interfaces provide essential granular control for some users. The idea is rather to ensure that internal product abstractions do not become an obstacle for the user.

Trust then becomes a central element. When users are no longer required to understand the underlying concepts, they also lose the ability to audit them. Trust must therefore be built and maintained through solid structures, safeguards, rigorous evaluation frameworks, and feedback loops to detect errors. AI must communicate its actions in terms of concrete and reversible outcomes.

The Future of AI in Products

Today, most AIs integrated into products focus on the interaction layer, enhancing speed and efficiency. However, the real opportunity lies in reinventing the concepts that users must understand to use a product. This involves re-evaluating each concept and determining whether it is truly necessary for the user or if it is an internal abstraction that AI could manage.

AI must extend beyond mere interaction to reach the conceptual layer. This is where it can truly reduce the learning threshold and transform the user experience.