Initially, Roberts intended to have the switching done by the mainframes that ARPA wanted to connect. But small, speedy minicomputers were just then appearing, and an adviser, Wesley Clark of Washington University in St. Louis, persuaded him to assign one of them to each of the research centers as a switch. Unlike the mainframes, which came from a variety of manufacturers, these so-called interface message processors, or IMPs, could have standardized routing software, which would save on programming costs and allow easy upgrades. In early 1969 the job of building and operating the network was awarded to the consulting firm of Bolt Beranek and Newman, Inc. (BBN), in Cambridge, Massachusetts. Although modest in size, BBN employed a stellar cast of engineers and scientists, drawn largely from nearby Harvard University and MIT.

Roberts had outlined what the IMPs would do. First, they would break data from a host mainframe into packets of about 1,000 bits each, attaching source and destination information to each packet, along with digits used to check for transmission errors. The IMPs would then choose optimal routes for the individual packets and reassemble the message at the other end. All the traffic would flow on leased telephone lines that could handle 50,000 bits per second. The BBN team, led by Robert Kahn of MIT, worked out the details and devised an implementation strategy. ARPANET was up and running at four sites by late 1969. At first, just four time-sharing computers were connected, but more hosts and nodes quickly followed, and the network was further expanded by reconfiguring the IMPs so they could accept data from small terminals as well as mainframes.

The nature of the traffic was not what ARPA had expected, however. As time went on, the computer scientists on the network used it primarily for personal communication rather than resource sharing. The first program for sending electronic mail from one computer to another was written in 1972—almost on a whim—by Ray Tomlinson, an engineer at BBN. He earned a kind of alphanumerical immortality in the process. For his addressing format he needed a symbol to clearly separate names from computer locations. He looked at the keyboard in front of him and made a swift choice: "The one that was most obvious was the @ sign, because this person was @ this other computer," he later explained. "At the time, there was nobody with an @ sign in their name that I was aware of." Trillions of e-mails would be stamped accordingly.