For roughly a decade, the question of which protocols the world’s computers would speak was genuinely open. On one side stood the Open Systems Interconnection suite, the official answer: a seven-layer reference model standardized by the International Organization for Standardization, backed by national telecom monopolies, and endorsed by governments. The United States even issued a procurement profile, GOSIP, that was supposed to require federal agencies to buy OSI products. On the other side stood TCP/IP, an unofficial research protocol from the ARPANET community with no comparable institutional blessing. By most political logic, OSI should have won. It did not.
The clash was visible in the technical literature years before it was settled. RFC 871, “A Perspective on the ARPANET Reference Model” by Michael Padlipsky of MITRE, dated September 1982, explicitly contrasts the ARPANET reference model with the ISO model. Padlipsky argues that the ARPANET model treats layers as broad functional groupings rather than a rigid seven-layer mandate, and he mocks the OSI requirement to traverse every layer with a memorable image: in the strict model you must climb down the stairs and then back up just to visit a neighbor on the same floor. His decisive point was practical: the ARPANET protocols were already implemented and running, while the OSI protocols were still being specified.
That practical advantage hardened into a culture. The internet community standardized in the open through RFCs and rough running prototypes, while OSI was produced by formal international committees that negotiated elaborate specifications before much code existed. The contrast was crystallized in a July 1992 plenary talk at the 24th IETF meeting, where David Clark of MIT presented “A Cloudy Crystal Ball: Visions of the Future.” His slides contained the line that became the internet engineering community’s creed: the rejection of “kings, presidents, and voting,” and a belief in “rough consensus and running code.” The phrase captured exactly why TCP/IP was winning. Decisions followed working implementations, not the other way around.
While the committees deliberated, the running code kept spreading. TCP/IP shipped in Berkeley Unix, ran on the ARPANET after the 1983 flag day, and scaled across NSFNET to thousands and then millions of machines. OSI products existed and were sometimes mandated, but they were late, expensive, complex, and interoperated poorly, and the network everyone could actually join was the TCP/IP internet. Governments that had mandated OSI quietly relaxed those mandates; GOSIP faded. By the mid-1990s the contest was effectively over.
OSI is the canonical dead end of internet history, and its failure is instructive rather than merely embarrassing. The seven-layer model it produced survives as a teaching tool and a shared vocabulary, so the effort was not wasted. But as a deployed protocol suite it lost, and it lost to a competitor with far less authority behind it. The lesson the internet community drew, and still repeats, is that in networking the thing that runs and that people can build on tends to beat the thing that is merely official.