How New Heuristics Are Reshaping the Creative Process Between Humans and Machines

When the wave of generative AI tools began flooding the market, I must confess my reaction was mixed: a sense of fascination for the possibilities and concern for the ethical challenges looming on the horizon.
As a digital product designer for over a decade, I had already witnessed several technological revolutions — but none with the transformative potential of generative AI.

Amid this whirlwind of innovation, I found guidance when I came across the work of André Neves, a researcher from the Federal University of Pernambuco, on “Heuristics for the Evaluation and Design of Generative AI Applications.”
It felt like finding a detailed map while navigating uncharted waters. His framework not only organized my chaotic thoughts on the subject, but fundamentally transformed my approach to designing AI-powered products.

Generative AI in Design

Before diving into the heuristics themselves, it’s worth contextualizing what we are experiencing.
Generative AI is not just another tool — it represents a paradigm shift in the relationship between humans and computers. Systems like ChatGPT, Claude, Midjourney, and DALL·E are not just assistants; they are collaborators with creative capabilities of their own.

This subtle yet profound change calls for a complete rethinking of our design methodologies.
We are no longer designing passive tools, but rather platforms for creative partnerships between humans and machines. And it’s within this context that Neves’ heuristics become critically relevant.

The Six Dimensions That Revolutionized My Design Process

The framework proposed by Neves is structured around six fundamental principles, unfolding into 24 practical heuristics. Each principle illuminates a crucial aspect of human-AI interaction, forming a holistic foundation to tackle this new design challenge.

1. Responsible Design: Ethics from the First Sketch

In past projects, ethical considerations often came too late — after many key decisions were already made. Neves’ heuristics taught me to reverse that logic.

Recently, while designing a virtual assistant for a fintech, I incorporated ethical checkpoints from the very first ideation sessions. We created usage scenarios that deliberately tested the system’s ethical boundaries: What if someone asked for potentially harmful financial advice? How would it handle sensitive data? How can we protect vulnerable users?

This shift in mindset not only produced a more ethical product but also saved countless hours of rework that would have been required to fix ethical issues later on.

2. Design for Mental Models: Navigating Expectations

The heuristics related to mental models were particularly enlightening. I realized much of the user frustration with generative AI systems stems from a mismatch between expectations and actual performance.

In a recent project for an AI-powered educational app, we conducted extensive research to understand how teachers conceptualized “intelligence” in AI. Many expected behavior similar to a human assistant, including consistent memory and contextual understanding.

This led us to completely redesign the conversational interface, adding visual cues that subtly conveyed the system’s capabilities and limitations. For example, we included “active context” indicators that showed which parts of the conversation were being considered — a visual way to explain the system’s functioning without overwhelming users with technical detail.

3. Appropriate Trust and Dependence: The Delicate Balance

One of the most dangerous pitfalls in generative AI systems is what researchers call “automation bias” — when users either trust the system blindly or, conversely, develop complete mistrust.

Following Neves’ heuristics, I implemented what I call “contextual education” in my projects — small, in-flow interventions that help calibrate user trust. For instance, in a content creation assistant, the system proactively identifies areas where its response is likely to be inaccurate and flags that to the user.

The result was surprising: users not only developed more appropriate levels of trust but also used the system more effectively, complementing its limitations with human checks when necessary.

4. Generative Variability: Turning Inconsistency into an Advantage

The inherent variability of generative systems initially seemed like a problem to be minimized. Neves’ heuristics helped me reframe it as a feature that could become a competitive advantage.

In a creative writing assistant project, we deliberately embraced variability by adding controls that let users adjust the “degree of surprise” in generated suggestions. Users could choose between more predictable, consistent outputs or opt for bold, diverse ideas.

This approach not only improved user satisfaction but significantly expanded the product’s use cases. Experienced writers often chose higher variability for inspiration, while beginners preferred more guided and conservative suggestions.

5. Co-Creation: The Dance Between Human and Machine

The co-creation heuristics were perhaps the most transformative for my practice. They challenged me to completely rethink the interaction model — from “human commands, machine obeys” to a genuine partnership.

In a recent tool for designers, we implemented what we called “mutual inspiration loops.” The system offered initial suggestions, users refined or rejected specific elements, and the system learned from those interactions to generate better-aligned proposals. The result was an experience that felt truly collaborative.

User feedback reflected this shift: instead of instrumental descriptions (“the tool helped me…”), we began hearing partnership narratives (“we worked together to…”). This subtle change in language signaled a profound transformation in the user-system relationship.

6. Imperfection: The Art of Failing Gracefully

Perhaps the most counterintuitive principle for me was embracing imperfection. As designers, we are trained to seek perfection, eliminate errors, and create flawless experiences.

Neves’ heuristics taught me that with generative AI, this approach can be counterproductive. Instead, I began designing for “graceful failure” — systems that not only minimize mistakes but also fail in helpful and transparent ways.

In an academic research assistant project, we implemented what we called “progressive transparency.” When the system detected low confidence in its response, it not only alerted the user but also exposed its internal reasoning and consulted sources. Users reported that these “transparent failures” were often more helpful than seemingly perfect answers, as they allowed them to evaluate the information independently.

From Theory to Practice: Turning Heuristics into Methodology

The true value of Neves’ heuristics emerges when they are methodologically integrated into the design process. At my studio, we developed a practical framework that embeds these heuristics at every stage:

• Discovery Phase: We use the heuristics as lenses to analyze user needs and market opportunities. For example, we specifically map how the co-creation principle can unlock new use cases.
• Ideation Phase: We structure brainstorming sessions around the six principles to ensure all dimensions are considered from the start.
• Prototyping: We develop specific tests for each heuristic cluster to validate whether the design meets the established criteria.
• Evaluation: We create an evaluation matrix based on the 24 heuristics, enabling a systematic and comparative analysis of different design approaches.

Putting It into Practice: ChatGPT vs. Claude

One of the most revealing analyses I conducted was comparing the ChatGPT and Claude interfaces using Neves’ heuristics. This analysis revealed fascinating design patterns:

ChatGPT excelled in heuristics related to accessibility and ease of use, with a minimalist and intuitive interface. However, it fell short on transparency and user education heuristics, rarely explaining its limitations or reasoning process.

Claude, on the other hand, showed greater attention to responsible design and transparency, often contextualizing its answers and clearly stating limitations. However, it sometimes sacrificed user experience smoothness for the sake of precision.

These observations not only informed my own design process but also provided valuable insights into different philosophies behind conversational AI design.

A New Horizon for Human-Centered Design

As generative AI becomes an increasingly present force in our digital lives, the heuristics proposed by Neves represent more than a set of guidelines — they are an invitation to fundamentally rethink what human-centered design means in the age of artificial intelligence.

In my personal journey, these heuristics evolved from analytical tools to guiding principles that permeate my entire creative process. They reminded me that even in a world where machines can generate creative content, the role of the human designer remains crucial — not just as a creator, but as an ethical guardian and facilitator of truly productive human-machine partnerships.

For those interested in diving deeper into this topic, I highly recommend reading André Neves’ original article, available on his LinkedIn.
And to those, like me, who navigate the turbulent waters of AI design daily, I hope these reflections offer both inspiration and practical guidance.

The generative AI revolution has only just begun, and together, we are defining not just products, but the very nature of our relationship with technology in the years to come. May we do so with wisdom, empathy, and vision.

References

Neves, A. M. M. (2024). Fundamental Assumptions of Design. LinkedIn. Retrieved April 18, 2025, from https://pt.linkedin.com/pulse/pressupostos-fundamentais-do-design-andr%C3%A9-neves-bmndf

Mariquito, G. (2024). How Generative AI Algorithms Can Help Interaction Designers Design Inclusive Interfaces During the Ideation Phase. ResearchGate. Retrieved April 18, 2025, from https://www.researchgate.net/publication/381769434

Neves, A. M. M., & Meira, V. (2023). Seminar Debates the Implications and Analytical Models of Generative AI. Institute of Advanced Studies, USP. Retrieved April 18, 2025, from https://www.iea.usp.br/noticias/inteligencias-individuais-sociais-artificiais

Generative AI and Its Approach to Design was originally published in UX Planet on Medium, where people are continuing the conversation by highlighting and responding to this story.