The history of sales technology, by year: from CRM to agents

What it solved: memory. Before CRM, the customer list lived in a rep's Rolodex and left when they did. ACT! put contacts on a desktop in 1987, Siebel Systems turned that into enterprise software in the 1990s, and Salesforce, founded in 1999, moved it to the browser as a subscription. HubSpot, founded in 2006, brought the same idea to inbound marketing.

Contact data followed. LinkedIn launched in 2003 and became the de facto directory of who works where. Jigsaw crowdsourced business contacts and became Salesforce's Data.com after a roughly $142 million acquisition. DiscoverOrg, founded in 2007, built a researched contact database and later took the ZoomInfo name.

What it broke: the CRM became a system of record reps had to feed, not a tool that helped them sell. Data entry became the job nobody wanted, and data quality has been the problem ever since.

What it solved: sending. Once reps had lists, the bottleneck was reaching enough people. Salesloft (2011) and Outreach (2014) turned outreach into sequences: a cadence of emails, calls, and LinkedIn touches that ran on a schedule. Apollo, which started as ZenProspect in 2015, bundled a contact database with a sequencer and later grew into one of the most used prospecting tools.

Outreach vs. Salesloft became the defining rivalry of the era. Both sold sequencing, dialing, and analytics to SDR teams by the seat, and both raised at their highest valuations in 2021: Outreach at a $4.4 billion valuation, Salesloft with a $2.3 billion Vista investment. In 2024 Salesloft bought Drift, and in 2025 it agreed to merge with Clari, a sign that standalone sequencing had become a feature.

What it broke: the inbox. Making it easy to send thousands of emails meant thousands of companies did. Volume went up, reply rates went down, and email providers tightened their filters. Cold email infrastructure became its own category as a result.

What it solved: coaching and forecasting. Gong and Chorus.ai, both founded in 2015, recorded and transcribed sales calls so managers could hear what actually happened instead of reading notes. Clari applied the same idea to the forecast. Gong was valued at $7.25 billion in 2021, and ZoomInfo bought Chorus for about $575 million the same year.

What it broke: nothing badly, but it showed the limit of dashboards. Teams had more insight than they could act on. Knowing that top reps ask more discovery questions did not tell anyone who to call tomorrow. More on today's options in call recording software.

What it solved: targeting. With sending and recording solved, the question became who to contact and when. Intent data from Bombora and 6sense, job-change tracking, website visitor identification, and hiring signals all answered "why now". Clay, founded in 2017, let an operator pull from dozens of data providers in one table, run waterfalls, and add AI research per row. It reached a $7.1 billion valuation in 2026.

Unify, founded in 2023, built on the same idea from the other side: plays triggered by intent that queue up outreach for reps. Amplemarket and Common Room approached signals from sales engagement and community data. Disclosure: I joined Unify as employee #9 in 2024 and ran its growth with a team of two, running 40+ plays a day across 100+ sequences.

This is the era that created the GTM engineer: someone who could connect data providers, signals, scoring, and the sequencer into one system. The full list of signals worth tracking is in the buying signals guide.

What it broke: simplicity and cost control. A modern stack has a CRM, a sequencer, a data workbench, three signal feeds, and a dozen providers, priced in credits, seats, and contracts that renew on different dates. Many teams ended up with enriched data and no pipeline.

What it solves: labor. ChatGPT launched on November 30, 2022, and within two years the question changed from "which tool" to "can an agent do this". Two very different answers emerged.

The first was the AI SDR: an agent sold as a replacement for a sales rep. 11x, founded in 2022, raised a $50 million Series B in 2024, and Artisan put up "Stop Hiring Humans" billboards in San Francisco. In March 2025 TechCrunch reported that 11x had listed companies as customers that said they were not, which 11x disputed in part, and its founder stepped down as CEO that May. Fully autonomous selling is still unproven. My take on the category is in AI SDR tools.

The second answer was agentic tooling for the operator. Coding agents like Cursor and Claude Code (generally available since May 2025) made it practical to build GTM pipelines by describing them. A new set of GTM tools now sells a CLI or API for those agents rather than a UI: Deepline lets coding agents run enrichment, signals, and CRM updates as "plays"; Freckle pitches itself as a Clay alternative for GTM engineers who work in coding agents; and OXYGEN combines tables, sequencing, an AI CRM, and agents behind a CLI and MCP server. Disclosure: OneGTM builds most new systems on Deepline and Claude Code.

This is what people mean by agentic sales: agents doing the reading, research, data work, and first drafts across the funnel, triggered by an instruction instead of a click.

I build GTM systems for a living, so this is a view from the work, not a forecast.

Seats give way to instructions. The unit of software used to be a seat for a person clicking through a UI. More and more, it is an API call an agent makes on someone's behalf, priced by usage. Tools that only work through their own screens will feel slow.

The GTM engineer becomes the default owner. Someone has to decide which signals matter, write the scoring, guard the send path, and check the agents' work. That is a GTM engineering job, and it grows as the tools get easier, because more can be built.

Data quality matters more, not less. Agents act on whatever they are given at machine speed. A wrong email or a stale job title used to waste one rep's afternoon. Now it can go out ten thousand times.

People stay on the conversation. The work that has held up best is agents doing the reading and preparation while a person approves what reaches a buyer. I expect that line to move, slowly, as trust is earned per workflow.

Niche data wins. When every team can run the same waterfall, the edge moves to data nobody packages: registries, permits, filings, and signals specific to one market.