The gap between the promise and the commercial success of knowledge-driven applications is still a ravine, and Gary Hendrix, developer of the natural-language capability of Symantec’s Q&A database manager word processor, had some advice for attendees at the recent IEEE AI Applications Conference in Orlando, Florida. Hendrix, who is director of research at the Cupertino, California company, and holds a PhD in artificial intelligence, stated that knowledge systems engineers should think of themselves as developers not of artificial intelligence programs but of business productivity software. Hendrix doesn’t give a damn whether Q&A has real artificial intelligence or ‘artificial’ artificial intelligence – as long as it understands English. He also cautioned researchers to be especially wary of demos – they can be designed to be so impressive that even the developer of the software begins to believe his own PR. He revealed that the basic natural-language feature of Q&A is a descendant of a 1975 Lisp program called Ladder; the part of developing Q&A was designing a product that would be useful and easy to sell. With that in mind, Symantec patterned Q&A after the PFS: family from Software Publishing because most computer salespeople already knew how to sell that software. A sizable amount of effort was also spent designing Q&A to fit in a micro’s memory and to perform at acceptable speeds. He responded to the common complaint that the natural-language feature of Q&A is too slow by pointing out that version 2.0, released last October, is four times faster and that another version due later this year would be faster still.