Ray Tomlinson forgot the words of the first network email

The setup

Ray Tomlinson worked at Bolt Beranek and Newman on TENEX, a time-sharing operating system for Digital PDP-10 computers, when he adapted an existing local mail program called SNDMSG to send messages across the ARPANET. In his own account of the work, Tomlinson describes folding in code from an experimental file-transfer program, CPYNET, so that SNDMSG could append a message to a mailbox file on a remote machine, not only a local one, closing the gap between existing single-computer mail and the network he had access to.

What the documents show

Tomlinson's account states he chose the at sign to separate a local user name from a remote host because the character already meant "at," as in a unit price, and needed no new meaning invented for it. The Internet Hall of Fame's inductee record, compiled independently of Tomlinson's own page, corroborates the same mechanism: SNDMSG modified to route mail through a network connection, with the at sign drawn from the Teletype's punctuation keys to build the user-at-host format that became the standard. Both records place the work in 1971, with the capability folded into TENEX's next general release in early 1972.

The friction

Tomlinson's own account is unusually candid about what it cannot establish: he describes sending test messages between two machines sitting side by side, connected only through the ARPANET, and says the messages were entirely forgettable and that he had, therefore, forgotten them, guessing the first was something like "QWERTYUIOP." He does not claim a specific date, only late 1971. That leaves the exact day, and the literal first words sent over network email, permanently undocumented by the one person positioned to know them.

What changed in the work

What the documents support is a specific, bounded claim: a working method for addressing mail across machines, built by combining two existing local tools, with a naming convention that persisted essentially unchanged for decades. What they do not support is any origin myth about a meaningful first message; the record explicitly denies one exists to recover. The setup cost of that first network mail was, by Tomlinson's telling, a small code change and an already-common character, not a redesigned system, worth remembering given how much infrastructure, spam filtering, threading, retention policy, has since attached itself to that same addressing convention.

• Does a tool's own documentation distinguish a confirmed date from a "late" or "circa" approximation, the way Tomlinson's account does?
• What happens to a workflow's founding story when the original record turns out to be this thin?
• How much of the current cost of running email, security, compliance, deliverability, was absent from the addressing problem Tomlinson actually solved?

The most reliable fact in this record is also the least dramatic one: a naming convention survived essentially untouched from an unremembered test message to a communication system now layered with authentication, compliance and AI drafting tools. The gap between that modest origin and the infrastructure built on top of it is itself worth noticing.