Designing AI Interfaces

Chapter 6. Agentic AI: Designing for Systems That Plan, Act, and Adapt

In 1997, Microsoft introduced Clippit, the animated paper clip assistant better known as Clippy. Intended to act as a friendly assistant, Clippy would intrude whenever it seemed like a user was lost. “It looks like you’re writing a letter,” it might say. “Would you like help?”

Microsoft’s vision for Clippy derived from research from Stanford University showing that people tend to treat computers like social beings. This suggested that if computers could behave more like helpful colleagues, users would accept and engage with them. Microsoft had tested this premise in 1995 with Microsoft Bob, a friendly interface featuring the dog Rover, designed to help people with minimal computer literacy navigate their PCs without the need for a manual (see Figure 6-1). However, Bob flopped within a year, with critics complaining that the slow-running tool infantilized users.

Undeterred, Microsoft hired an illustrator to design an expanded cast of potential assistants for Office 97. The resulting characters numbered over 250 and included animals, robots, globes, professors modeled on Shakespeare and Einstein, even office supplies like a stapler. The paper clip emerged as a frontrunner, and Microsoft built it into Office’s help system.

Clippy’s technology was somewhat simple, powered by basic heuristics and Bayesian algorithms that monitored user actions for recognizable patterns. On paper, the idea made sense: Clippy would ...