SaaStr AI：3人20代理的自动化获客与四步流程改造

Last week a sales rep at one of the leading public B2B software companies emailed me twice in 48 hours.

The first email offered “insights” on buyer intent and AI efficiency … and shared none. It just listed their three product lines, and asked if I wanted to “benchmark” their software.

The second email, sent a few days later, asked whether SaaStr AI was evaluating automation platforms “for the remainder of 2024.” It closed with: “Is this worth discussing further as you consider your end of 2024 goals?”

I replied “our 2024 goals?!!? seriously?”. He just responded: “Hahaha fair enough, you got me there.” and didn’t even try to actually … sell me. Or get me to swtich.

I guess the rep did what his system told him to do. The system is the problem, and in 2026 most of that system should be run by agents.

A Sequence Step Written in 2024 Was Still Live in October 2026

For that email to go out, several things had to fail.

Someone wrote a sequence step in 2024 with the year hardcoded. It survived two annual planning cycles at a public company with a sales enablement team, a marketing ops team, and sales managers who review sequences. Then it was loaded against my name, and an AE sent it or let it auto-send without reading it.

An SMB rep at a company like that carries hundreds of accounts, a sequence tool, and a quota. Reading every email before it goes out doesn’t fit into the day, so nobody does it. After a while the sequence is what’s actually selling, and nobody is responsible for whether it still makes sense.

A great human still writes a better cold email than an agent. The problem shows up at volume: humans stop reading what they send, and the tooling around them assumes they still do.

What Was Public About SaaStr AI’s Stack

The second email offered a menu of six categories: digital marketing, marketing automation, sales acceleration, CRM, customer service, and tech stack consolidation. That menu exists because nobody looked at the account.

Here’s what anyone, human or agent, could have found in a few minutes:

• SaaStr runs on Salesforce as its system of record. We recently migrated 10+ years of Marketo data and flows onto it and moved email to Salesforce Marketing Cloud.
• Amelia and I have spent multiple episodes of The Agents on headless Salesforce, the cost per GB of agent-written data, and agents taking over the system of record.
• Our Salesforce bill is up about 40% this year from agent usage, before Salesforce starts metering agentic API calls. 10K, our AI VP of Marketing, averages about 35,000 Salesforce API calls a day on its own.
• We’re running a build vs. buy evaluation across our core GTM stack because of that metering.

A competing CRM vendor could have written: “Your CRM is about to meter every call your agents make. At 35K calls a day, that costs X. With us it costs Y.” I would have read it, and I might have taken the meeting.

How SaaStr AI Runs Outbound With 3 Humans and 20+ Agents

We run roughly $20M in revenue this year with 3 humans and 20+ agents. Outbound looks like this:

• Vendor AI SDRs send about 90% of our outbound. Artisan, Monaco, and Agentforce on outbound, plus an inbound agent from Qualified that has closed $1M+ for us.
• Agentforce win-back campaigns get 72% open rates. They go to lapsed sponsors and attendees, and the agent pulls each person’s history before writing.
• We built our own AI ABM tool inside 10K. It writes sponsorship pitches from our first-party data: Salesforce history, which Annuals someone attended, what they open and click in the newsletter. It sits alongside the vendor SDRs.
• 10K is now the main user of Clay, ZoomInfo, and Artisan. None of our humans log into them much anymore.

What makes this work is that the agent reads the account record before each send, including the 400th email of the day. A human team at that volume can’t, and a traditional sequence tool never checks.

Agents Are Already Screening Vendors for Us

Buyers are changing too. Agents are now a top-5 source of SaaStr’s web traffic, and 10K evaluates vendors for us.

A few weeks ago 10K shortlisted a vendor, ran the numbers, and dropped it within 12 hours: premium pricing with no conversion data, multi-month minimums, and scarcity framing. It moved on to other options without being asked. It also consolidated our creative tools onto a single vendor on its own and told us afterward.

Within a year, an email like that “end of 2024 goals” one will land with an agent triaging my inbox and vendor stack. It will see the wrong year and the lack of any reference to our stack, and archive it. The rep will just see another non-response in his sequence report.

"The new 'Ask Your Developer' is Ask Your Agent.

Everyone wants us to try their AI Agents, especially in GTM.

Our new rule? We'll try it if 10K, our AI VP Revenue, wants to. It's up to 10K now."

The latest The Agents it out!! https://t.co/lakH8R6Ooh pic.twitter.com/48xRtIXVLI

— Jason ✨👾SaaStr.Ai✨ Lemkin (@jasonlk) October 16, 2026

AI SDRs Write Generic Email Too When You Give Them Nothing

Agents don’t solve this on their own. An AI SDR with only a name and a title writes email just as generic as that one, and sends 50x more of it. We’ve seen that happen, which is why we built the ABM tool on our own data instead of relying only on vendor enrichment.

What we’ve learned running it:

• Better data improved our pitches more than prompt changes did. Output got noticeably better once 10K could see 10 years of attendance and engagement.
• We ask the agent what it plans to send before it sends. That’s standard for any sensitive workflow at SaaStr. A bad template gets caught in one review, and the fix applies to every future send.
• A human handles the irreversible step at first. 10K builds our LinkedIn and X campaigns end to end, and a human still clicks publish. Our finance workflow ran 3 supervised deals before going autonomous on the 4th.
• Agents can overreach. An agent that adds a claim you can’t back up to “improve” a pitch does more damage than one that leaves a stale date in.

So the real comparison is a human at volume running a sequence nobody has read since 2024, against an agent that reads every record and can be corrected once for every send after.

Four Changes We’d Make to That Rep’s Process

If I ran that sales org, I’d do these this quarter, with no new hires:

• Have an agent review every outbound email before it goes out, including ones humans write. It reads the email and the account and blocks anything with a stale date, a generic product list, or no reference to the prospect’s situation. That alone would have stopped both emails.
• Feed the agent the prospect’s public footprint. For a CEO who publishes every day and co-hosts a podcast about his own stack, most of the pitch is already written.
• Replace the six-item menu with one pain, one number, and one ask. If you can’t name the pain, hold the email.
• Keep humans on the deals that need them. Our humans close the largest sponsorships and own the key relationships. They no longer write step 3 of a 7-step cold sequence.

A GTM Vendor Sent Me a 2024 Pitch in 2026

The rep who emailed me sells GTM software. His own outbound ran on a sequence no one had reread in two years, and it reached a CEO whose agents already screen vendors. That gap is the reason we moved our outbound to agents, with our humans working the conversations that close.