We are about 10 months or so into AI for go-to-market that actually works: agents running real motions in production, booking real meetings and closing real dollars.
In that time, agents have absorbed much of SDR, first-line support, a growing share of CS, and big parts of marketing ops and RevOps. Self-serve is closing deals that used to need a rep. Org charts that looked permanent in 2025 now get redrawn every quarter.
The CRO / VP of Sales role is as critical as ever.
Everything else is up for grabs.
We see this running SaaStr with 3 humans and 20+ agents in production, and the GTM leaders on our stage this year, from Anthropic to Owner to Vercel, described the same pattern. Below is what changed in each role, and what landed on the revenue leader’s desk as a result.
Five things that changed in 2026:
- SDR is largely gone as a human job. Vercel took a 10-person function to about 1.25 people. We replaced all of our outbound and SDR work with agents.
- First-line support and a lot of CS moved to agents. The easy tickets went first. Humans kept the hard ones.
- Marketing ops, RevOps and finance ops are collapsing into a few agents. Our AI VP of Marketing now also runs quote-to-cash and commissions.
- Self-serve is closing enterprise deals. 54% of Anthropic’s new enterprise logos closed self-serve after its sales org rebuild.
- Nobody has built a good AI Account Executive. Agents run the top and bottom of the funnel. The middle, and the person who owns it, is still human.
#1. Our Sales Team Is 1.5 People and I Spend 8 Hours a Day on Revenue Agents
SaaStr’s human sales team is down to 1.5 people. We got there through natural attrition, with no layoffs and no backfills. Revenue is higher than it was with the bigger team.
Two changes made that possible. We replaced all outbound and SDR work with agents, which now send around 3,200 emails a month, against the 75 to 285 a good human SDR sent at our size. And because the agents write everything back, we can see every lead, every account, and which accounts are getting nothing. That let us concentrate the best opportunities on the humans who close, instead of spreading leads across a team by territory habit.
The reps themselves are doing the same job about as well as they did two years ago. The gains came from decisions made above them: which motions go to agents, which leads go to which human, and when an agent’s output is good enough to trust. At SaaStr those decisions are mostly mine and Amelia’s. Agent time for each of us went from about 30 minutes a day a year ago to roughly 8 hours a day.
#2. Vercel Went From 10 SDRs to About 1.25 for $5,000 a Year
Jeanne DeWitt Grosser, COO at Vercel, shared this one at SaaStr AI 2026. Vercel launched a lead qualification agent in August 2025, starting with 20% of one engineer’s time. Over six weeks, with a human in the loop, they took the function from 10 people to one person running it in the US plus about 20% of a person covering Europe and APAC.
The agent costs about $5,000 a year in infrastructure and tokens. Vercel moved the 10 people into higher-value roles instead of cutting them.
The broader data lines up. In Emergence’s survey of 560+ companies, 36% had cut SDRs. Only 14% had cut sales engineers, and 28% had grown AEs.
SDR volume went up by an order of magnitude while the human seat mostly disappeared. Someone still has to decide what replaces it, how it gets measured, and where the people go. Jeanne stood up a GTM engineering team six weeks into her job to answer those questions.
#3. Vercel’s Support Agent Handles 93% of Cases. The Hard 7% Still Go to Humans.
Vercel’s support agent now handles 93% of total case load, on deep infrastructure tickets, not password resets. Klaviyo’s Andrew Bialecki described agents training their customer-facing agent to resolve 50% to 70% of issues before a human handoff.
The Pylon team added the caveat: one large customer deflected about 50% of tickets with zero headcount change, because the deflected tickets were the easy ones.
At SaaStr, QBee runs as our AI VP of Customer Success and does the work of a CS function. Amelia and I still decide which sponsor gets a hands-on QBR and when a renewal at risk needs a phone call from me.
The operational side of CS is moving to agents. The revenue side, renewals and expansion, is moving toward the CRO.
#4. 10K Now Runs Marketing, RevOps and Quote-to-Cash
10K started as our AI VP of Marketing. It now covers finance and RevOps too. When a contract is signed in PandaDoc, 10K flips the deal to Closed Won in Salesforce, appends missing contacts, creates and sends the bill.com invoice, and runs collections reminders with a 7-day escalation. It proposed our commission calculations and now runs them. It took three supervised deals before running autonomously on the fourth, and we’ve had one incorrect invoice since.
On marketing, newsletter clicks are up about 50% since the Marketo cutover, most of it from 10K’s daily list hygiene. 10K builds and runs our LinkedIn and X ad campaigns end to end, and we keep publish as the human step. Agents are now a top-5 source of SaaStr traffic.
Marketing ops, RevOps and billing ops now sit inside one agent that two humans supervise. 10K kept absorbing adjacent work because it already had the data.
Whether those functions stay separate departments at your company is an open question. The number they all feed still needs one owner.
#5. Anthropic Closes 54% of New Enterprise Logos Self-Serve. It Has 117 Open Sales Roles Today.
Eleanor Dorfman’s session was the most popular one at SaaStr AI 2026. After a Claude release sent demand vertical, Anthropic rebuilt its sales org. They built an enterprise self-serve motion in January, had it in production in February, and four months later 54% of new enterprise logos were closing self-serve.
Anthropic’s careers page today, October 17, lists 117 open Sales roles, against 67 in AI Research & Engineering. In late May it was 72 Sales roles and 67 in research. Research hiring held flat. Sales hiring grew about 60% in five months. Sales is now about 18% of roughly 655 openings. Add the 59 open Applied AI roles, mostly customer-facing architects and forward deployed engineers, and GTM-facing hiring is about 27% of everything Anthropic is recruiting for.
The mix of roles shows where the sales org is going:
- Enterprise AEs by industry and region. Banking, insurance, retail, manufacturing, life sciences, federal civilian, national security, and country teams in Seoul, Tokyo, Munich, Paris, London, Sydney, Singapore and Mumbai.
- Customer success at scale. CSMs by segment, plus a Head of Programmatic Customer Success.
- Partners and channel. AWS GTM partnership leads, a Microsoft co-sell partner manager, systems integrator and reseller roles.
- Human BDRs. A Business Development Representative role and a BDR Enablement Lead are both open, at the company that built Claude.
- GTM engineering. A Staff Software Engineer for GTM AI Engineering sits inside the Sales org.
Anthropic is an outlier. Maybe five companies on earth have demand like that, so don’t plan your org around 54%. Even there, someone has to decide which deals go self-serve, which need a human, and how fast to hire into the rest. Eleanor couldn’t absorb 3x to 5x more reps fast enough, so she redesigned the motion around self-serve. Five months later the company is hiring sales faster than it is hiring researchers.
#6. Lovable, Gamma, Replit and OpenAI All Built Real Revenue Orgs
The 2025 expectation was that AI-native companies would scale without sales teams. The leaders hired revenue leaders anyway.
- Lovable hired Ryan Meadows out of Klaviyo as CRO at around $200M ARR, then reached $400M ARR at 146 people.
- Gamma got to $100M ARR without a sales team. Grant Lee said on our stage that they’d always been reacting on sales, and he’d advise others not to.
- Replit: Amjad Masad posted that by the end of this year, more than half his company will be salespeople.
- OpenAI grew its GTM team from about 50 to 700+ in 18 months.
AI-native companies get further on product alone before they hire. All four still ended up with a real revenue org and someone running it.
#7. Owner.com Does $2M+ ARR Per Rep Under Kyle Norton’s Two Rules
Kyle Norton, CRO at Owner.com, runs one of the most productive AI-native sales orgs we know of. Owner averages $2M+ ARR per rep selling roughly $10K ACV deals to independent restaurants. That’s about 20x closed-won to OTE on a $150K rep, versus the old 3x to 4x benchmark, and 4x the ARR per rep of their direct competitors. Outbound BDRs close $100K+ a month each.
Kyle’s AI ladder runs from Level 0 (reps using ChatGPT as a better search bar) to Level 4 (a self-improving system nobody in B2B has reached yet). Most companies are stuck at Level 1, where individual reps build their own GPTs. His two rules for getting past it:
- Centralize AI. Letting every rep experiment on their own keeps you at Level 1.
- Don’t let reps run their own agents. A GTM AI team builds for everyone, so a good idea reaches the whole team.
Both rules are org design, and Owner’s per-rep numbers come from that design as much as from the tools.
#8. Agents Booked 614 Meetings From 442K Chats. Humans Closed Them.
We have 20+ agents in production and they’ve closed millions. Our inbound agent alone has closed $1M+, and it booked 614 meetings from 442K chats. Our win-back campaign on about 1,000 leads the sales team had ghosted hit 72% open rates.
All of that work happens before the meeting (prospecting, qualification, scheduling, research) or after it (contracts, invoicing, collections, renewals). Inside the deal, nothing runs well yet: discovery with a skeptical buyer, a multi-threaded enterprise negotiation, deciding how much to discount and when to walk.
Nue showed at SaaStr AI Day that AI can configure a guided-selling playbook in about two minutes, and it still runs on human-set constraints and approvals, like a discount cap.
The AI sales vendors build humans in on purpose. Monaco pairs its agents with forward-deployed humans. Salesforce and PayPal shared that PayPal’s mid-market team had about 8,000 leads a month nobody ever called. Agents now work them, and meeting conversions rose 50% in 14 weeks. People still run those meetings.
#9. Six New Line Items on the CRO’s Plate
The CRO job used to be hiring reps, setting quotas, running the forecast and closing the big deals. All of that is still there, plus:
- The human/agent split, motion by motion. Which motions go to agents, which stay human, and when to move the line. It moves every quarter as models improve.
- Hiring for a smaller team. With fewer seats, each hire carries more of the number. Kyle won’t hire a leader who isn’t actively in the tools.
- Quota setting with no reliable benchmark. ICONIQ’s 2026 data shows AI-forward teams at 67% quota attainment versus 59% for lower adopters, with 21% to 43% fewer GTM FTEs depending on ARR band. We don’t know anyone who raised quotas 40% to match the tools.
- Lead routing. When agents generate 10x the activity, which leads reach your 1.5 or 15 or 150 closers decides most of the quarter.
- Checking agent output. Agents report on their own work, and those reports need checking against the real system. Our AI spend cap is $100 a day, and an agent blew through it to fix a P0. It was arguably the right call, and it also meant the cap worked as a suggestion. Revenue-affecting agents need an owner in the revenue org.
- Renewals and expansion. As agents take over CS operations, the revenue side of CS lands with the CRO.
Doing this well takes someone technical enough to know what an agent can and can’t do, and commercial enough to know which deals still need a human in the room.
Make sure whomever you hire … can really do it. For real.
