Soon, however, harried plant managers may get some help from so- called intelligent agents--software emissaries to cyberspace. Equipped with just enough smarts and knowledge to carry out assigned tasks, such virtual aides have already been harnessed for such simple things as enabling Amazon.com Inc. to suggest new books based on a person' s purchases.

ZIPPING AROUND. In manufacturing, however, agents offer remarkable benefits. ``If you structure them correctly, they can do lots of neat things,'' says engineer Howell Mitchell of Flavors Technology Inc. in Manchester, N. H. For instance, Mitchell thinks agents could one day spiff up production lines. The basic idea: Equip each factory robot with a software agent. Then, if the robot conks out, the agent zips around the plant's intranet to find another machine that can take over the task. When it does, it triggers a reprogramming of the assembly line--and production hums along.

The agents could also be given bigger tasks. Venturing out into the larger world of cyberspace, they could automatically seek out suppliers, buying parts and raw materials precisely when needed. They would even be made ``smart'' enough to do such things as pick the supplier with the fastest delivery time in an emergency. Indeed, programmers foresee that there will be whole menageries of agents--production line agents, buyers and sellers, ``middle'' agents to facilitate transactions, even ``reputation'' agents to help rank suppliers by quality. ``A great deal of business will be mediated automatically by billions of software agents doing the same kinds of transactions humans do now,'' says Steve White, a senior manager at IBM Research.

With more agile assembly lines, proponents predict improvements in plant efficiency of up to 30%. Parts and materials costs should drop because agents can find suppliers all over the world, not just the handful in the procurement manager's Rolodex. And with agent-assisted factories able to reconfigure production lines on the fly, ``we should see drastically reduced cycle times and more mass customization,'' says Neil Christopher, program manager at the National Institute of Standards & Technology. That's why everyone from Boeing and General Motors to Intel and Deere is exploring the use of agents at various levels of manufacturing.

``UNCHARTED WATERS.'' Don't expect intelligent agents to revolutionize manufacturing just yet. Flavors' Mitchell learned that agents aren' t yet capable of rearranging an assembly line to walk around a broken robot, for instance--the reprogramming is too difficult. Martin Hill, head of e-business development in Britain for Sweden's Intentia International, points out that agents really aren't very adaptive. They may be clever enough to pick a supplier with the best price, but they don't yet have the brains to spot those that offer innovative solutions. ``There, the idea breaks down a bit,'' he says.

What's more, an industry in which decisions are made by millions of pieces of software ``is exciting, but it's totally uncharted waters, '' says Michael Jeng-Ping Shaw, a business professor of the University of Illinois. When White and others at IBM had agents compete to sell goods, they created a never-ending cycle of widely fluctuating prices. ``That's not a great solution for consumers,'' says White. For now, intelligent agents have a lot of potential--but also a lot to prove.