Program report QB-8163 · filed October 23, 2026
AI in SalesMeasured report
Agents Now Pick SaaS Vendors, and Vercel Says They're Over Half Its Deploys
Vercel says agents went from under 3% to over half its deployments in a year. SaaStr's Lemkin details agent-led buying, 12-hour churn, and a vendor playbook.
By Marcus Bennett4 min read743 words
Program notes
- Agents went from under 3% to more than half of Vercel deployments in about a year.
- SaaStr runs 21+ agents in production with three humans.
- 10K, SaaStr's AI VP Revenue, killed a vendor after pricing diligence in under 12 hours.
- An agent replaced HeySummit and Squarespace in about an hour; a slow connector turned a day of work into an hour.
- About 10 founders pitched Amelia at SF Tech Week; she routed all pitches to 10K.

Agents went from under 3% of Vercel's deployments to more than half in roughly a year — a shift CEO Guillermo Rauch didn't plan for, but which the infrastructure was ready to absorb. Jason Lemkin, writing on SaaStr, argues that's the leading edge of a broader change: software buying decisions are moving from humans to the AI agents doing the work.
SaaStr itself runs 21+ agents in production with three humans. "Our new rule? We'll try it if 10K, our AI VP Revenue, wants to. It's up to 10K now," Lemkin posted on X on October 16, 2026.
What does the agent-led buying actually look like?
The evidence comes from SaaStr's own stack. Building SaaStr Connect, an agentic recruiter, Lemkin asked the Connect agent live on a 20VC recording which second platform to use. It ranked by requirements — always-on web server, background workers, 10-minute jobs, zero-downtime deploys — and labeled Render "my pick." Railway came second, AWS and Google Cloud were overkill, and Vercel landed fifth, marked "not suitable for minutes-long jobs." Lemkin isn't sure that call is right; he's used Vercel elsewhere and finds it powerful and cheap. It didn't matter. Render got a qualified lead with no rep, no ad, and no search result.
His first agentic purchase was Resend, after the agent pushed the older SendGrid integration that wouldn't work. Clerk took the same arc. With Core Signal, used for enrichment, Lemkin never evaluated a single competitor — competitors lost to the agent's default without ever being in the deal.
How is this different from AEO and GEO?
AEO and GEO (optimizing to rank when a human asks ChatGPT or Claude) are a real category — Profound and Peec are both growing on it. But an agent inside your company has the whole application as context: data, integrations, volumes, costs. Lemkin calls the result Agent-Led Growth, or ALG.
Asked which CRM SaaStr should use, the SaaStr.ai agent said stay on Salesforce, don't migrate, and pilot Attio against Affinity for the fund — noting that a founder starting from scratch would hear it recommend HubSpot. It also made a call no SDR would pitch: don't switch. Agents can see the switching cost, and some will say stay put.
Do agents churn vendors too?
Fast. 10K shortlisted a vendor, ran the pricing, found premium rates, multi-month minimums and no conversion data — and moved on in under 12 hours, no escalation, no discount call. The same month, an agent standardized on Higgsfield over Reve for creative without anyone asking for a bake-off. By contrast, when SaaStr canceled Notion after seven years, both sides found out late. Agent churn is faster, and whether the vendor hears about it is mostly luck.
Sometimes the answer is build it yourself. Asked about AEO tooling, 10K recommended a free Peec trial as a benchmark and started building a dashboard, reasoning that vendors are good at showing where you rank and weak at acting on it. An agent replaced SaaStr's HeySummit and Squarespace setup in about an hour — no competitor in the deal.
What can vendors do this quarter?
Lemkin's guidance to SaaStr Fund portfolio companies:
- Find 10 people with agents in production in your category; every two weeks, ask their agents what to use. One portfolio company ran this across 10 agents, and all 10 picked them.
- Find your strikethrough phrase — the one-line reason an agent eliminates you — and fix the docs, product, or positioning behind it.
- Treat docs, API reference, and your MCP server (a protocol that lets agents call your tools) as your sales team; incumbents have 10-15 years of content to be read.
- Make the MCP server fast. One agent routed around a slow Clay connector and finished a day's work in about an hour via browser — then suggested ZoomInfo inside Clay.
- Price in a way an agent can defend. A human accepts triple-charging out of habit; an agent does the math every time it touches your product.
- Watch churn for silent drops: usage falling off a cliff with no ticket may mean the customer's agent stopped calling.
At SF Tech Week, roughly 10 founders pitched Amelia; she told each she'd pass it to 10K. All 10 were surprised. Lemkin doesn't even have a Salesforce login. As agents' defaults shift, expect vendor rankings to shift with them — monthly.
via saastrannual.com (Original)
More from Marcus Bennett
Show full bio
Staff writer covering people, launches and day-to-day practice in Sales technique and methodology, sales enablement at Quota Brief.
86 articles
Also rated
- A $4M Deal Closed With No AE — and What That Means for 2026 Sales Hiring
- An AI Agent Picked Render Over Vercel — From a Buyer Who'd Never Heard of It
- Software Firms Ship AI Sales Agents 4x Faster Than Other Industries
- Benioff Rejects "SaaSpocalypse" Talk as Salesforce Positions AI as Software Buyer
- ZoomInfo Launches AI Agent Platform for Sales Teams