2001 to 2010: email becomes infrastructure, and the filters learn rules
Mailchimp and SendGrid made sending at scale a service. Blocklists, scoring rules, Bayesian filters and the CAN-SPAM Act answered. The pattern that still runs outbound in 2026 started here.
Jared Hawkins-Young · Founder, Lost Star TechnologyOctober 8, 20266 min readPart 1 of 14Key takeaways
- Between 2001 and 2010 sending email at scale stopped being an engineering project. Mailchimp (2001) made lists and templates self-service; SendGrid (2009) made sending an API.
- Mailbox providers answered with blocklists, rule-based scoring and, from 2002, Bayesian filters that learned from what people marked as spam. Gmail (2004) trained that learning at scale.
- The CAN-SPAM Act (effective January 1, 2004) set the rules senders still follow, and did not require consent. The inbox, not the law, became the gatekeeper.
- The pattern that repeats for the next twenty-five years starts here: every tool that makes sending easier hands the inbox a new pattern to defend against.
You inherited the outbound motion along with the title. A domain, a sequencer, a list someone bought in 2024, and a reply rate nobody can explain. Before you touch any of it, it helps to know that every rule you now work under was written as a reaction to a tool that made sending easier. This series walks that history era by era, then lays out the operating model that follows from it. It starts in 2001.
Sending stopped being an engineering project
In 2001 Mailchimp launched as a side project inside a web design agency and made list management, templates and bulk sends self-service for anyone with a browser. Before that, sending a few thousand emails meant renting servers, configuring mail transfer agents and keeping a developer close by. In 2009 SendGrid turned sending into an API, so a product could push transactional and marketing mail at scale without running mail infrastructure at all.
By the end of the decade the marginal cost of one more email was close to zero, and the capability that once needed an operations team was a monthly subscription. That is the gold column in the timeline below. It is also the only column most outbound playbooks ever talk about.
The inbox learned rules
Filters of that era relied on patterns they had already seen. Blocklists such as Spamhaus, founded in 1998, published the addresses and networks known to send unwanted mail, and providers refused connections from them outright. SpamAssassin, first released in 2001, scored every message against hundreds of rules: suspicious phrasing, forged headers, too many links, too much colour. Cross a threshold and the message went to the junk folder.
In August 2002 Paul Graham published A Plan for Spam, which popularised Bayesian filtering: instead of a fixed rulebook, the filter learned word probabilities from the mail each person marked as spam. When Gmail launched in April 2004 it brought that kind of learning to a provider with the user base to train it at scale. The defense had stopped being a list of rules written by hand. It had become a model.
Law arrived in the same window. The CAN-SPAM Act took effect on January 1, 2004 and set the rules commercial senders in the United States still follow: honest headers and subject lines, a physical postal address, a working opt-out honoured within ten business days. It did not require consent before the first email. That is why cold email is legal in the US, and why the inbox, not the law, became the real gatekeeper.
The pattern, stated once
- Part 12001 to 2010Sending technologyCommercial email infrastructureSelf-serviceSend APIsBulk listsMailbox defenseRule-based spam filteringRulesBlocklistsContent scoring
- Part 22011 to 2017Sending technologySequencing turns follow-up into a systemSequencesSales engagementFollow-upMailbox defenseAdaptive filtering and sender reputationReputationEngagementAdaptive filters
- Part 32018 to 2021Sending technologyData connects to executionEnrichmentCRM syncSignalsMailbox defenseIdentity becomes a stronger signalDomain identityAlignmentFeedback
- Part 42021 to 2023Sending technologyMulti-mailbox scale meets generative draftingMailbox rotationWarmupAI draftingMailbox defensePattern recognition learns fasterBehaviorInfrastructureFeedback loops
- Part 5Feb 2024 to 2025Sending technologyMore throughput, more identity fragmentationOrchestrationAI researchMore capacityMailbox defense · formal enforcementAuthentication becomes a hard gateSPF · DKIM · DMARCOne-click opt-outSpam < 0.3%
- Part 62026 to 2027Sender capabilityResearch and drafting become nearly freeAgent researchDynamic variantsReply routingInbox prioritizationVisibility becomes the next gateRelationship contextMessage contextPriority visibility
Every breakthrough that made outreach easier gave mailbox providers a new pattern to defend against. Bulk sending produced blocklists. Templates produced content scoring. The same shape repeats in every era that follows: sequencers meet sender reputation, enrichment meets identity checks, generative drafting meets machine classification, mailbox rotation meets mandatory authentication.
If you take one idea from this series, take this one. The channel is shared infrastructure, and the party that decides whether you may use it is not you. Everything in the operating model later in the series, from treating a domain as production infrastructure to a person approving every send, follows from that.
What era one means for a head of sales in 2026
- Reputation is cumulative. The blocklists of 2001 never went away. They sit under the machine learning layers that came later, and a domain that lands on one carries that history into every campaign after it.
- The rules outlived the tools. Nobody runs the 2004 version of Gmail, but the CAN-SPAM requirements are unchanged, and the one-click opt-out that Google and Yahoo made mandatory in 2024 is their direct descendant.
- Volume was the first advantage and the first thing to stop being one. By 2010 anyone could send ten thousand emails. The question had already moved to whether any of them should be sent.
Next in the series
Part 2 picks up in 2011, when Outreach, Reply and Apollo turned follow-up into a configurable system, and Gmail started grading senders by their history rather than by the message in front of it. The series page lists every part in order.
Common questions
- Was cold email legal in the United States in 2004, and is it now?
- Yes on both counts. The CAN-SPAM Act regulates commercial email without requiring prior consent: honest headers and subject lines, a physical address, a working opt-out honoured within ten business days, and no sending to anyone who has opted out. Those requirements are unchanged in 2026. What changed is enforcement by the mailbox providers themselves, covered in part 5.
- Why does a 2001 filter matter if I use a modern sequencer?
- Because the early layers never went away. Blocklists and content scoring still run underneath the machine-learning filters that came later. A domain that trips them carries that reputation into every later campaign, whatever tool sends the mail.
Sources
- Mailchimp, company history (founded 2001)
- Twilio SendGrid, about (founded 2009)
- The Spamhaus Project, about (founded 1998)
- Apache SpamAssassin, project history (first release 2001)
- Paul Graham, A Plan for Spam, August 2002
- Google, Gmail launch announcement, April 1, 2004
- FTC, CAN-SPAM Act: a compliance guide for business (effective January 1, 2004)
Written by Jared Hawkins-Young from the system he builds and runs. Diagrams are drawn from the same specs as the software. Published October 8, 2026. Illustrative numbers are labelled as such; every other figure carries its source.