The era of designers acting as "helicopter parents" to AI is over; the future demands a "character coach" approach, according to Craig Wattrus, AI Design Engineer at Flatfile. Speaking at the AI Engineer World's Fair, Wattrus outlined Flatfile's innovative approach to user experience design for AI, emphasizing a collaborative paradigm where humans and AI work in concert, rather than humans simply commanding machines. This shift fundamentally rethinks how we interact with intelligent systems, moving beyond traditional interfaces to embrace AI's unique capabilities and emergent behaviors.
Wattrus presented Flatfile's AI stack through four distinct UI paradigms: invisible, ambient, inline, and conversational. Flatfile, a developer platform focused on data migration and transformation, has found fertile ground for AI integration, leveraging large language models' code-writing abilities. The invisible AI, for instance, dynamically generates personalized demos for new users. "You don't need to know that AI is working on it," Wattrus explained, highlighting how the system intuitively crafts a Flatfile application perfectly suited to a user's specific industry or use case in the background.
Ambient AI functions by subtly analyzing data for potential improvements, indicated by "sparkles" that highlight opportunities for correction. More direct interaction comes with inline AI, where users actively work within data, and the AI agent proposes and executes transformations. These agents are not merely suggesting fixes; "These agents are writing code that then gets run on this data set," Wattrus noted, enabling rapid, large-scale data cleansing and standardization with impressive efficiency. This allows for millions of rows to be processed quickly, eliminating much of the manual effort traditionally associated with data preparation.
The conversational paradigm, familiar through chatbots, is where the "character coach" metaphor truly shines. Wattrus recounted his realization that he had been "giving engineers feedback on our agents like, ‘oh it shouldn’t start saying this,’ and ‘it shouldn’t use these words.’" He understood that he was treating the AI like a human, trying to control its every output. Instead, the goal should be to "build out the nature" of the AI, guiding its personality and behavior rather than dictating its every move. Tools like a "chat tuner" allow designers to adjust parameters such as tone, proactiveness, and cautiousness, observing how these changes impact the AI's interactions.
This iterative process of "feeling the material" of AI—understanding its inherent properties, strengths, and limitations—is paramount. Just as a woodworker must understand the grain of wood, AI designers must deeply comprehend how models function to create effective "form factors." Wattrus demonstrated "AI cursors" in Figma, where AI could autonomously build UI elements based on prompts. While initially exciting, he quickly realized this constrained the AI, likening it to "putting a Formula 1 driver in a Prius." The goal is not to force AI into human-like tools but to develop interfaces that allow AI to excel in its own way.
The ultimate aim is "courting emergence," moving beyond simply automating tedious tasks. Flatfile's product, Obvious, exemplifies this by intelligently combining disparate datasets and generating comprehensive reports, identifying duplicates, and suggesting next steps without explicit instruction. Wattrus highlighted an instance where the AI, unable to directly fix missing employee IDs, suggested the human user contact HR, demonstrating a nuanced understanding of workflow and human-AI collaboration. "I wasn’t expecting it to be able to help the human to go and do the job where it couldn’t," he remarked, underscoring the unexpected value that emerges when AI is allowed to leverage its unique capabilities. This approach champions a future where AI isn't just a tool, but a true coworker, contributing meaningfully and autonomously within a well-defined operational framework.
