The role of a designer at an AI-native company shifts from traditional interface design to defining success metrics. This involves evaluating the quality and format of AI-generated content and deciding on the heuristics for what constitutes a good output, making it a categorically different job.
The creative superpower of LLMs isn't perfection, but massive volume. By embracing the "thousand pots" approach, teams can generate a vast quantity of ideas in minutes. This deluge of content, while mostly unusable, will contain a few inspiring nuggets for iteration that human-only processes might miss.
Because AI tools allow teams to build and iterate so quickly, the "fear of time" diminishes. This reduces the need for traditional planning artifacts like roadmaps and presentation decks, allowing teams to focus on building and discussing tangible work in the present moment.
In fast-paced AI development, traditional specializations like "designer" and "engineer" merge. Individuals contribute based on their strengths rather than a fixed job description, with everyone empowered to ship code and design mockups, creating a more equalized playing field where specialization feels vague.
Evaluating AI model outputs ("evals") is a new, crucial part of the product development cycle. It acts as a form of synthetic user research, but its primary challenge is anticipating the wide variety of user prompts and avoiding the pitfall of optimizing for a narrow, biased set of inputs.
When designing for a massive, diverse user base, traditional persona-based HCI methods fail. The focus must shift to creating simple, general mechanics and then rigorously testing how they succeed and, more importantly, how they fail at their absolute extremes. Understand where the app "declares bankruptcy."
While AI tools can get you from zero to a functioning prototype in a weekend, the "second 80%" of the work—navigating dependencies, rewriting for different environments, and dealing with production complexities—is still a grueling, multi-week process. The path to shipping remains as difficult as ever.
In a rapidly changing environment, formal roadmaps create friction. OpenAI's alternative is a clear company-wide "rallying cry"—the one or two most important goals. This empowers teams to self-organize and swarm priorities collaboratively, reducing the internal strife caused by competing priorities.
In traditional design sprints, teams vote on static ideas due to the high cost of prototyping. With LLMs, you can skip voting. Instead, generate multiple rich, interactive prototypes for the top ideas simultaneously, allowing decisions based on tangible user experiences, not abstract concepts on sticky notes.