Roberts started the design for the network using the networking theory Len Kleinrock had developed in his thesis, Information Flow in Large Communication Nets.
In early 1967 Roberts met with some of the principal investigators of the Intergalactic Network to begin determining the requirements for the network. Kleinrock, the network analysics specialist, wanted there to be diagnostic tools at every node. The time-sharing expert at UC Berkeley, Herb Baskin, demanded a response time of no more than half a second.
At a 1967 Ann Arbor computer conference, Roberts presented his initial plans to the Intergalactic Network, meeting skepticism but also getting the crucial idea for the IMP from Wes Clark. Then, at a 1967 Gatlinburg computer conference Roberts presented his plans for an "ARPAnet", and learned about the work of Donald Davies and Paul Baran from Roger Scantlebury. In August 1968 Roberts sent the ARPAnet IMP RFP to 140 companies. BBN was awarded the contract to build the IMP as Christmas Day approached.
BBN shipped the first IMP August 30, 1969 to Kleinrock's group at UCLA, and the next IMP to Engelbart's team at SRI on October 1. On October 29, the first message was sent over the nascent network. The researchers at UCLA tried to send the message "LOGIN" one character at a time. They successfully sent "LO" before the SRI host crashed. They soon were able to get the computers successfully communicating, the beginning of the ARPAnet.