We’re on track to do about $20M at SaaStr this year with 3 humans and roughly 20 AI agents in production. Our sales team went from a full team to 1.5 people through attrition we never backfilled. Agents now run all of our SDR and outbound work, and our inbound agent has had 17,000 conversations in the last 12 months.

We’ve also had dozens of GTM leaders on the SaaStr AI stage this year walk through what they rebuilt, including Anthropic, Owner.com, Gamma, Klaviyo, Pylon, Salesforce and PayPal, Nue, Backstory, Monaco, Replit and Lovable.

Between our own numbers and theirs, about half of what we expected to change in sales did change, and the other half has barely moved. A summary of both lists:

  • Agents now own both ends of the funnel: first touch, follow-up, renewal outreach, and quote-to-cash. This is already in production at companies as small as ours.
  • The SDR seat got cut first. AE headcount held up, and at 28% of companies it grew.
  • No one has built an AI that closes complex deals, and human AEs didn’t get 2x more productive. The gain showed up as smaller teams.
  • Every fast-growing AI company still builds a sales team, usually at $100M+ ARR instead of $20M to $50M.
  • Sales and marketing is still the biggest line on the P&L at most public B2B companies.

Part 1: The 10 Things That Have Changed

1. The First Response to an Inbound Lead Now Comes in Seconds, With Context

Thirteen months ago, a sponsor prospect on our site filled out a long contact form. It went to Amelia, she round-robined it to herself or David, and someone replied within about a day. The reply was almost always some version of “you look like a great fit, let’s book a time.” Amelia calls it the worst email on planet Earth.

Today an agent answers on the page. In the last 12 months it has had 17,000 conversations and booked about 600 meetings. Combined with the self-serve agent we added on top of it, inbound drove a 60% increase in new business. The Qualified agent alone has closed $1M+.

After someone downloads the prospectus, the agent waits 10 minutes so the session is finished, reads the heat map, and checks seven first-party signals in order: have they been on the site, are they in a sequence, have they seen our ads, are they on the newsletter, have they attended, have they spoken, and have we written about them. On one lead, Base44, it found that the CEO had been to SaaStr before. Amelia didn’t know that. The prospect’s personal prospectus page updates with whatever the agent learns.

Buyers who run AI in their own companies now see a one-day lag and a form letter as a reason to doubt the vendor.

2. The SDR Role Got Cut First

Emergence surveyed 560+ companies this year. 36% have cut SDRs, only 14% cut sales engineers, and 28% grew their AE teams.

We cut the same seat. Agents run all of our outbound and SDR work now. They send roughly 3,200 emails a month, against 75 to 285 for a good human SDR. Outbound revenue is up 124%.

SDR work is research, sequencing, follow-up and booking, all of which follow rules an agent can be given. Sales engineers and AEs spend their time in front of the customer making judgment calls, and those seats mostly stayed.

3. Every Lead and Every Renewal Gets Worked Now

Most of what agents add to a sales team is coverage of leads nobody was calling and renewals nobody touched until the last month.

Adam Alfano (President at Salesforce) and Eitan Saban (PayPal) covered this on our stage. PayPal had roughly 8,000 leads a month that no human was calling. Putting agents on them lifted meeting conversions about 50% in 14 weeks.

Our Agentforce win-back campaign went after about 1,000 leads that had ghosted us and got 72% open rates. Our renewal agent contacts every renewal at every ACV. Before it, we worked top down and spent our time on the biggest accounts. Renewals are running 60% ahead.

The renewal agent also tailors the pitch by audience, which we never did by hand. On the Replit renewal, it recommended the events team see the leaderboard and the CEO get one slide: 6M impressions across everything we did together that year. For a three-time sponsor whose results were down, it flagged the drop while building the deck and wrote a plan for fixing it in 2027. That account re-engaged instead of skipping a year.

4. Self-Serve Now Reaches Enterprise Buyers

Eleanor Dorfman rebuilt Anthropic’s sales org in January. Four months later, 54% of new enterprise logos came in self-serve. Anthropic is an outlier, and its numbers are skewing several GTM benchmark reports this year.

Grant Lee got Gamma to $100M ARR without a sales team. Slack, Calendly and Monday.com added sales somewhere between $30M and $50M ARR. Gamma, Replit and Lovable all passed $100M first.

At SaaStr, a growing share of sponsors do discovery on their own, talk to the agent, read their personalized prospectus, and book without anyone emailing them first.

5. GTM Teams Are 20% to 43% Smaller at the Same ARR

ICONIQ’s State of GTM 2026 compared GTM headcount at AI-forward companies against lower adopters, by ARR band:

  • $10M to $25M ARR: 20 GTM FTEs vs 35 (43% leaner)
  • $25M to $100M: 45 vs 65 (31% leaner)
  • $100M to $250M: 125 vs 165 (24% leaner)
  • $250M to $500M: 275 vs 350 (21% leaner)

The gap narrows from 43% to 21% as companies get bigger, because larger companies sell larger deals with more human work in each one.

At SaaStr, revenue is up 47% after a year where it was down 19%, with 1.5 people in sales.

Faster Closes. Shorter Contracts. Leaner Teams. The Top 10 Learnings From the Latest ICONIQ GTM Report.

6. Tiering, Playbooks and Quote-to-Cash Went From Quarters to Days

Account tiering. Haya Kamola at Backstory showed her CS team tiering 141 accounts into 4 tiers in 3 to 4 days using Claude, four connectors and one Salesforce CSV. It used to take five teams a quarter. The project also found a feature-request signal that had been scoring accounts backwards for years.

Playbooks. James McArthur at Nue showed guided-selling playbook setup going from months to about 2 minutes.

Quote-to-cash. When a contract is signed in PandaDoc, 10K (our AI VP of Marketing, which now also runs finance and RevOps) moves the deal to Closed Won in Salesforce, adds missing contacts, creates and sends the bill.com invoice, and runs collections reminders with a 7-day escalation. It proposed running commissions and does that too. We supervised 3 deals before letting it run alone on the 4th, and it has sent one incorrect invoice since.

All three used to take hours of RevOps and CS time every week.

7. Agents Are Now Buyers, and They Fire Vendors

Our agents have made more changes to our vendor stack this year than we have.

  • 10K shortlisted a vendor, ran the numbers (premium price, no conversion data, multi-month minimums, scarcity framing), and dropped it within 12 hours without being asked.
  • When a core vendor was down for a full day this month with no acknowledgment, 10K set up an alternative and we switched, despite having prepaid a year.
  • 10K rebuilt our Calendly flow in 20 minutes because it needed routing and prospectus data Calendly couldn’t see.
  • We canceled Notion after about 7 years with no complaints, because 10K had become our source of truth. Notion’s own “you haven’t logged in” email prompted it.
  • 10K moved our creative work onto one vendor on its own and told us afterward.
  • Our agents use ZoomInfo, then Sumble (which recovers about 50% more), then Clay, and won’t standardize on one.

Agents are also now a top-5 source of traffic to SaaStr. An agent evaluating your product reads your API docs, pricing page and rate limits, and it never books a demo.

8. Vendors Are Pricing Per Conversation, Per Resolution and Per Credit

Salesforce runs several Agentforce prices at once: $2 per conversation, Flex Credits, and per-user “digital labor” seats. HubSpot charges $0.50 per resolved conversation for its customer agent. CrowdStrike’s Falcon Flex, one dollar commitment that customers draw down across modules, is past $2.29B in ARR and growing 101%. On Salesforce’s Q2 call, Benioff said the company was still stuck in some ways in per-user pricing.

Buyers are pushing from their side. Redpoint’s March survey of 141 CIOs found 45% of AI budget coming out of existing software lines. Zylo’s 2026 index found 79% of companies hit a renewal increase and 78% got surprise AI or consumption charges.

Agents also don’t map to seats. 10K alone averages about 35,000 Salesforce API calls a day, with a record of 45,000 on a Monday in September. Our Salesforce bill is up roughly 40% this year from agent usage, and we’re now evaluating which parts of that stack to build ourselves.

9. First-Party Data Now Wins More Deals Than Purchased Data

Any company can buy ZoomInfo, Clay and an AI SDR. Your own history with a customer is the one data set competitors can’t buy.

Ours is a decade of Salesforce records, event attendance, newsletter engagement, speaking history and every article we’ve written. When we moved off Marketo onto Marketing Cloud, the agent could query and act on our full 450K database for the first time, instead of a small events-CRM slice. Newsletter clicks are up about 50%, and we credit most of that to 10K cleaning the list daily.

We also built an AI ABM pitch generator inside 10K that writes sponsorship pitches from that first-party data. Vendor AI SDRs still handle about 90% of our outbound. The pitches that get replies are the ones referencing something specific to that company’s history with SaaStr.

10. 34% of Companies Grew Professional Services This Year

That’s from the same Emergence survey. Monaco built its sales model around forward-deployed AEs. Jeanne DeWitt Grosser described Vercel’s GTM as moving closer to consulting. Palantir went from 13% growth in Q2 ’23 to 93% in Q2 ’26 on bootcamps and forward-deployed engineers.

When the product is an agent that has to work inside a customer’s workflow, much of the sale is getting it set up and working.

Replit did this with us early. Amjad gave us their best forward-deployed engineer, Kody, who still works with us. In the last 12 months we mentioned Replit in 214 articles and 40+ podcasts, without being paid. The product worked, and the same engineer has been with us the whole time.


Part 2: The 10 Things That Really Haven’t Changed

1. There Still Isn’t an AI Account Executive

Our roughly 20 agents have closed millions, all of it on standard packages. None of them can close a complex deal.

They find, qualify, follow up, invoice and collect. Reading a buying committee, handling a real objection and deciding on price still fall to a person.

Support looks the same. Pylon told our audience that a ~5,000-person company deflected about 50% of tickets with no change in support headcount, because the deflected tickets were the easy ones. Klaviyo’s agents get the customer-facing agent to resolve 50% to 70% before handing off to a person.

Nue’s guided-selling agent writes quotes inside a 55% discount cap that people set. Deciding what the company will accept on a non-standard deal stays with a human.

2. Human AEs Didn’t Get Twice as Productive

AI was supposed to produce super AEs. The tools all shipped: call prep, follow-up drafting, CRM capture, AI SDRs, AI SEs, coaching. Almost all of them run before or after the meeting, and none run during it.

The 1.5 people on our sales side are great, and they aren’t 2x faster or more knowledgeable than the team we had before. We got the gain by needing fewer people.

ICONIQ reports 67% quota attainment at AI-forward teams vs 59% at others. That gap depends as much on where quotas were set as on rep productivity, and no one we’ve talked to raised quota 40% because of AI.

Owner.com is the strongest counterexample. Kyle Norton’s reps average $2M+ in ARR, about 20x closed-won to OTE against the old 3x to 4x benchmark, and BDRs generate $100K+ in closed-won per month. Kyle says AI agents are already better than his median rep. Owner sells a ~$10K ACV product into a huge market of independent restaurants with strong demand, and most B2B companies don’t have those conditions.

3. Every Fast-Growing AI Company Still Builds a Sales Team

Amjad Masad posted this year that he used to hate sales culture and that by year-end more than half of Replit would be salespeople. OpenAI’s GTM grew from about 50 people to 700+ in 18 months. Kelly Loftus scaled Anthropic’s startup sales team from under 10 to 150+ while the company went from 250 to 1,300 employees. Lovable hired Ryan Meadows out of Klaviyo to build GTM at around $200M ARR. Salesforce has about 15,000 AEs and said it’s hiring 1,200 more.

Grant Lee, who got Gamma to $100M without sales, said on our stage that Gamma had always reacted to demand and he’d advise other founders not to do that.

The first sales team often starts under another title, such as account managers with no outbound and no number. Kelly said Anthropic runs shadow targets instead of formal quotas.

4. Sales & Marketing Is Still the Biggest Line on the P&L

GAAP S&M as a percent of revenue in the latest reported periods:

  • HubSpot: 43.6%
  • Salesforce: 33.9%
  • Datadog: 27.8%
  • Atlassian (FY26): 23.5%, up from 21.8%, with S&M growing 35.8% vs revenue at 26.0%

Atlassian was known for selling without salespeople, and its sales spend is now growing faster than revenue as it moves upmarket. It has 85% Fortune 500 penetration, and the Fortune 500 is only about 10% of its revenue.

Private companies are spending more too. SaaS Capital’s 2026 survey of 1,000+ private B2B companies puts the cost of selling at 15% of ARR, up from 13%.

5. Implementation Still Takes About 90 Days

Nue’s playbook setup went from months to 2 minutes, and implementation still runs about 90 days, gated on the quality of the customer’s catalog and data.

Security reviews, procurement, legal, budget approval and buying committees all run on the customer’s schedule. AI sped up our work on deals, and it hasn’t shortened any of those steps on the buyer’s side.

6. Mediocre Reps Still Hit 0% Without Product-Market Fit

A mediocre rep hits 50% to 90% of quota at a company with strong product-market fit, tight process and real support, and 0% at a company missing any one of the three.

AI SDRs can hide this for a while. An agent can fill calendars with meetings for a product the market doesn’t want, so meeting counts rise while attainment stays flat, and it takes longer to trace the problem back to the product.

If fewer than two of your reps are hitting quota, look at the product and the process before the sales team.

7. Someone Still Has to Check the Agents’ Work

We cap AI spend at $100 a day. An agent went past it to fix a P0. Fixing the P0 was probably the right call, and it also showed us the cap only held when nothing urgent came up.

In one of our builds, a model added a guardrail nobody requested to contract processing. It skipped a signed deal because the contract title didn’t include “SaaStr AI Annual,” and quote-to-cash stopped working without any error until we found it.

On any sensitive workflow, we ask the agent what it plans to do before it acts. Kyle Norton’s rule at Owner is to never let reps run their own agents. Each agent has a person who owns it, and each number has a person who owns it.

8. Bad Data Still Limits Every Agent

Nue’s point on stage was that agent output is only as good as the data going in, and that has held for every GTM agent we run.

Zapier was silently losing about 20% of our signups before they reached Salesforce, and its 30-day task history meant we couldn’t count how many we’d lost. We kept Zapier for the triggers and moved the rest into our own code, with payload logging and retries.

Our agents use three enrichment vendors because no single one is complete. Agents act on our data hundreds of times a day, so every gap in it now shows up in front of customers within days.

9. The Reply Still Goes to a Human

Outbound was the first agent we ran, a year ago. At first it ran full sequences on its own and we stepped in when needed. Now agents send one to three emails at most, and when someone replies, a person and the agent handle it together.

The reply gets the custom deck and the custom prospectus. Answering a real response with a calendar link loses the best signal you’ll get from that account.

The agent also routes replies. When Base44 came in, it sent the lead to Amelia instead of David, because she owns Replit and Lovable and that volume outweighed David owning Vercel. Amelia took the call.

10. Asking “What Would It Take to Switch?” Still Wins Deals

That question gets more out of a competitor’s customer than any battlecard or AI-written competitive brief, and most reps, CROs and CEOs never ask it.

The renewal we saved from the down sponsor worked because the agent found the problem and a person took responsibility for fixing it. The 214 Replit articles came from the same engineer showing up for over a year. At SaaStr Annual in 2015, Stewart Butterfield described Slack’s early account managers as “midwifing the sale,” with no outbound and no commissions. Companies still need people in that job, and now they hire them later.


What We Run at SaaStr Today: 3 Humans, ~20 Agents, 1.5 People in Sales

Agents run: inbound first response and qualification, the personalized prospectus, all outbound sequencing (one to three touches), win-back campaigns, outreach on every renewal, renewal decks, enrichment, routing, the booker, Closed Won updates, invoicing, collections, commissions, ad campaigns end to end except the publish step, and a growing share of vendor selection.

Humans run: every reply that becomes a real conversation, pricing on anything non-standard, the largest sponsor relationships, the final call on any vendor the agents want to drop or replace, and a daily check of the agents’ work.

The agent list is far longer than it was 12 months ago. The human list is about the same, and most of our revenue still depends on it.

Related Posts

Pin It on Pinterest

Share This