Email: The Killer App That Predated the Web

Zusammenfassung

Email is the internet’s oldest surviving application and its most universally used — a system invented almost accidentally by a programmer at BBN Technologies in 1971 who wanted to see if he could send messages between machines on ARPANET. Within two years, email had become 75% of all ARPANET traffic. Through four decades of evolution — from Unix command-line clients to web browsers to mobile devices — it has outlasted every “email killer” successor. For the network it traveled on, see ARPANET: Building the Network That Became the Internet; for the privacy challenges it raised, see The Privacy War.

The Accident That Changed Communication

In the autumn of 1971, Ray Tomlinson was working at Bolt Beranek and Newman (BBN) in Cambridge, Massachusetts — the company that had built the IMPs that connected ARPANET’s first nodes. His official assignment was to work on TENEX, the time-sharing operating system BBN was developing for ARPANET hosts. As a side project, he was noodling with two programs.

The first was SNDMSG, a program that allowed users on a single time-sharing machine to leave messages for each other in electronic mailbox files. On a shared TENEX system, a user could type “SNDMSG” and compose a message to another user; the message would be appended to a file in the recipient’s directory. It was more convenient than leaving notes on desks, but it only worked within a single machine.

The second was CPYNET, a prototype file transfer program that could move files between different ARPANET hosts. It was experimental — not yet production software — but it worked: you could copy a file from a computer at UCLA to one at MIT.

Tomlinson had a lateral thought: SNDMSG wrote text to a mailbox file. CPYNET could write files to remote machines. What if SNDMSG could deposit its message into a mailbox file on a remote machine? He modified SNDMSG to use CPYNET’s network transfer capabilities when the recipient was on a different host, and the combination worked on the first try.

The only remaining design decision was notation: how should the address specify both the user and the host? Tomlinson looked at the keyboard and picked the @ symbol — already present, not used in usernames (which might contain letters, numbers, and periods, but not @), and semantically clear: “user AT host.” The choice was made in perhaps thirty seconds.

Tomlinson later said he couldn’t remember what the first email message said. In interviews, he guessed it was something like “QWERTYUIOP” — a test to confirm the mechanism worked, not a message worth preserving. He mentioned it to a colleague, who said “Don’t tell anyone — this isn’t what we’re supposed to be working on.” Nobody stopped him.

Why Email Took Over ARPANET

The network had been designed to solve a resource-sharing problem: computers at different institutions should be able to use each other’s computing capacity and access shared databases. The designers expected the dominant use to be remote login and batch computation. What they got was something they had not anticipated: people wanted to talk to each other.

Asynchronous communication turned out to be the revolutionary feature. A phone call required both parties to be available simultaneously. A memo required physical delivery. Email required neither: you sent it when you had something to say, and the recipient read it when they were ready. For a research community scattered across a dozen university campuses and military research centers, the ability to send a message at 2am knowing it would be waiting when your colleague arrived at 9am the next morning was genuinely transformative.

By 1973, email accounted for approximately 75% of all traffic on ARPANET — a statistic that astonished the network’s designers and prompted a DARPA-commissioned study of the phenomenon. The study found that scientists and researchers were using email not just for logistics but for intellectual exchange, draft-sharing, and the kind of informal conversation that usually happened in hallways. The network had become a new kind of social space.

The first email standards codified what Tomlinson’s ad-hoc invention had established. RFC 733 (1977) formalized the message format. RFC 822 (1982, Dave Crocker) refined it further, establishing the header format — “From:”, “To:”, “Subject:”, “Date:” — that survives to this day. Every email client in existence, from Outlook to Gmail to a Unix command line, still formats messages according to RFC 822’s structure.

sendmail and the Unix Email Ecosystem

For the internet’s first decade and a half, email was a Unix-native phenomenon. The dominant mail transfer agent was sendmail, written by Eric Allman at UC Berkeley beginning in 1981. Sendmail’s job was to accept email on a Unix server, figure out the routing to deliver it to its destination (consulting DNS’s MX records to find the recipient domain’s mail server), and hand it off to the next server in the chain.

Sendmail was extraordinarily capable and notoriously difficult to configure. Its configuration file — a text file of cryptic macros and rewriting rules — became legendary as one of the most impenetrable configuration systems in the history of software. An entire book, sendmail by Bryan Costales (O’Reilly, 1993), ran to nearly 800 pages and was widely considered inadequate. System administrators who could configure sendmail correctly commanded respect and job security.

Sendmail’s complexity also made it a security risk: its configuration surface was enormous, and its history included multiple significant vulnerabilities. The Morris Worm of 1988 exploited a debug backdoor in sendmail as one of its three attack vectors. Despite its problems, sendmail handled a majority of internet email for the better part of two decades and remained widely deployed into the 2000s.

The MX Record

Before email routing could work reliably at internet scale, DNS needed to be extended with a record type specifically for mail. The MX (Mail Exchange) record, introduced in 1983 alongside DNS itself, specifies which server handles email for a given domain. When a mail server wants to deliver a message to “user@example.com,” it looks up the MX record for “example.com” in DNS to find the address of example.com’s mail server. This separation — the mail server address is independent of the web server address for the same domain — is what allows email infrastructure to be managed separately from other internet services.

MIME: Teaching Email to Carry More Than Text

Original email was plain ASCII text — no formatting, no attachments, no non-English characters. RFC 822 explicitly required email content to be US-ASCII, which excluded not just binary files but also the accented characters needed for French, German, Spanish, and virtually every language except English.

MIME (Multipurpose Internet Mail Extensions), developed by Nathaniel Borenstein and Ned Freed and standardized in 1992, solved both problems. MIME defined a system for encoding binary data (images, documents, executables) as ASCII text for transmission in email bodies, and for specifying the encoding and character set of message content. A MIME email header declaring “Content-Type: image/jpeg” and “Content-Transfer-Encoding: base64” told the email client that the following blob of text was actually a JPEG image encoded in base64, and should be decoded and displayed accordingly.

MIME also enabled HTML email — messages formatted with fonts, colors, tables, and images. Purists objected strenuously (and some still do); the argument that email should be plain text for readability across all clients and accessibility tools has merit. But by the mid-1990s, HTML email had become standard for corporate communications and marketing, and plain-text email had retreated to technical communities and personal preference.

The Consumer Revolution: CompuServe, AOL, and Hotmail

For its first twenty years, email was the province of researchers, scientists, and corporate professionals with network connections. The consumer internet brought email to everyone, in three waves.

CompuServe and MCI Mail (both 1980s services) offered email to paying subscribers, though these were walled gardens: CompuServe users could email other CompuServe users, but connecting to the internet email system required workarounds. The user bases were modest — tens of thousands rather than millions.

America Online changed the scale. AOL’s email system, launched in 1993 with the service’s public expansion, introduced tens of millions of Americans to electronic mail. The famous “You’ve Got Mail” voice greeting — spoken by actor Elwood Edwards, recorded on a cassette tape in his living room in 1989 for the AOL predecessor service Q-Link — became a cultural touchstone. For many AOL subscribers, “email” and “AOL mail” were synonymous. The service also introduced many users to the concept of spam, as AOL’s mass marketing campaigns — including the infamous CD mailings that blanketed the US in the 1990s — accustomed users to unsolicited commercial messages.

Hotmail, launched on July 4, 1996 by Sabeer Bhatia and Jack Smith (the Independence Day launch date was deliberate and slightly corny), was the first major free web-based email service. The innovation was technical — email accessible from any browser, with no software to install — but the business model was the revolution: free accounts, subsidized by advertising. Hotmail grew to 1 million users in its first six months and 10 million by the end of 1997. Microsoft acquired it in December 1997 for approximately $400 million in stock, one of the defining acquisitions of the early internet era.

Gmail and the Modern Era

By 2004, the dominant web-based email services — Hotmail and Yahoo! Mail — offered users between 2 MB and 4 MB of storage. The unstated business model was storage scarcity: users regularly reached their limits, felt pressure to delete old messages, and remained engaged with the service to manage their quota. Email as an archive of conversations was not what these services wanted to enable.

On April 1, 2004 — a date chosen partly because the announcement seemed like an April Fool’s joke — Google launched Gmail with 1 GB of free storage. The differential was not incremental; it was a factor of 250. Gmail’s entire philosophy was that you should never need to delete email; you should search it instead. The interface was built around Google’s search technology, with threading that grouped related messages into conversations.

The storage offer was so far beyond anything competitors offered that journalists initially assumed it was an April Fool’s joke. It was not. Gmail’s storage advantage forced Yahoo! and Microsoft to dramatically increase their own limits within weeks, effectively ending the storage-scarcity model for web email.

Gmail was also technically innovative in ways less immediately visible: it used Ajax to update the inbox without full page reloads (unusual in 2004), introduced keyboard shortcuts inspired by vi and mutt, and built spam filtering that improved with use. By 2023, Gmail had over 1.8 billion active users.

The Spam Wars

Email’s openness — anyone could send a message to anyone, with no barrier to entry — made it the perfect vehicle for bulk advertising the moment it reached mass adoption. The first identifiable mass commercial email was sent by Gary Thuerk at Digital Equipment Corporation in 1978, advertising DEC’s new DECSYSTEM-20 to 393 ARPANET users. The recipients were furious; ARPANET’s acceptable use policy explicitly prohibited commercial use. Thuerk was reprimanded. The message reached its recipients anyway, and DEC sold some computers as a result.

The lesson was clear but ignored until the early 1990s, when internet access became commercially available and spam exploded. By the mid-1990s, bulk emailers were sending millions of messages daily; by 2004, spam represented over 90% of all email traffic globally. The economics were irresistible: sending a million emails cost almost nothing, and even a 0.01% conversion rate produced revenue.

The responses evolved in parallel with the problem. Rule-based filters blocked specific addresses and patterns. SpamAssassin (2001) combined dozens of heuristics into a scoring system. Bayesian filtering, popularized by Paul Graham’s 2002 essay “A Plan for Spam,” used statistical analysis of word frequencies to classify messages individually. These methods were partially effective but caught in an arms race with spammers who modified their messages to evade detection.

Legal approaches followed. The CAN-SPAM Act (2003) set requirements for commercial email in the US: accurate headers, a functional unsubscribe mechanism, no deceptive subject lines. The law was criticized as toothless — it effectively legalized commercial email while adding nominal constraints — and was largely ignored by the professional spamming operations that ran from Russia, China, and other jurisdictions beyond US reach.

Modern spam filtering rests on DKIM (DomainKeys Identified Mail), SPF (Sender Policy Framework), and DMARC — three complementary standards that together allow receiving servers to verify that an email actually came from the domain it claims to be from. These standards, widely deployed by the early 2010s, dramatically reduced forged-sender spam. Combined with machine learning classifiers, they have pushed spam to manageable levels in major email providers — though it remains the dominant category of email traffic by volume.

Email’s Strange Persistence

Since at least 2005, technology journalists and industry analysts have been announcing the death of email. Instant messaging would replace it (it didn’t). Facebook Messages would replace it (it didn’t). Slack would replace it (it didn’t). Microsoft Teams would replace it (it didn’t). The argument is always the same: email is too slow, too formal, too hard to search, too easily spammed, too fragmented between providers.

Email’s persistence despite these criticisms reflects a genuine technical virtue: it is the one universal communication protocol that works across organizational, national, and technical boundaries without requiring the recipient to be on the same platform as the sender. You can send email from Gmail to Outlook to a corporate mail server running Postfix on a Linux server in a university basement. No other messaging system achieves this interoperability. Slack messages cannot be sent to Teams users. iMessages cannot be received on Android without workarounds. Email has no such limitation.

This universality comes from its age and its open standards. Email’s format (RFC 5322, the successor to RFC 822) and transport protocol (SMTP, RFC 5321) are documented, implementable by anyone, and owned by no corporation. The cost of that openness is spam and the absence of end-to-end encryption in most deployments. The benefit is a communication system that no single company controls, cannot be shut down by any single entity, and continues to function regardless of which technology companies rise and fall.

Ray Tomlinson, who invented it in thirty minutes as a side project in 1971, died on March 5, 2016. He was 74. He had no idea, he said in interviews, that the @ symbol he chose that autumn afternoon would become the most recognized typographic character in the world.