Teamday
← Back

Vercel COO: A $5,000 Lead Agent and a 93% Support Agent

SaaStr AI
agentsenterprisebusinessautomationfuture-of-work

What Vercel's Internal AI Agents Actually Cost and Do

Jean Grosser is the COO of Vercel. About six weeks after joining, in June 2025, Grosser set up a go-to-market engineering team to bring agents into everything Vercel's go-to-market org does. In this SaaStr AI talk she gives the numbers after ten months. The customer support agent "now handles 93% of total caseload", even though many Vercel cases are highly technical. The content agent did 96% of major content updates last quarter. And the lead qualification agent launched in August 2025 took the sales development team for that function from 10 people to one, with the others moved to more valuable roles. (We covered the original 10-SDRs-to-1 story on Lenny's Podcast.)

The cost of the lead agent: The new detail is the bill. The lead agent "costs us under $5,000 a year between infrastructure and tokens to run" and takes about 20% of one engineer to maintain, which Grosser puts at a 32x ROI. One engineer prototyped it over a weekend and it reached production six weeks later, after running in shadow mode while Vercel's best SDR reviewed every output until she had nothing left to correct. The same approach now covers 30 SDR workflows, from event follow-up to time-based campaigns, and SDR quotas rose 30% this quarter.

A tripod for every agent: Each internal agent is built by three people: a GTM engineer, a data scientist and the best subject matter expert in the function. They document the expert's process, encode it as a tool-calling workflow before adding AI, and keep a human in the loop who QAs every action until the expert stops finding fixes. For the lead agent, the engineer watched every tab the SDR opened: LinkedIn, BuiltWith, the company website, the CRM and Slack history.

Headless agents in Slack: "Agents don't live in UIs. They call APIs." Vercel's meeting intelligence agent ingests every sales call, posts coaching notes to Slack, proposes CRM updates, tracks objections and runs a postmortem on every closed-lost deal. It only works because Gong has an accessible API and Salesforce has usable webhooks; Grosser says tools without developer surfaces get dropped. A playbook platform turns signals such as a usage spike or a pricing page visit into drafted outreach that reps approve with one click, so the best reps' plays run on every account.

d0 and the data foundation: The most used agent at Vercel is d0, a data analyst agent in Slack that turns natural language into SQL. Questions that took a week through a data team ticket now take under a minute. It runs on a semantic layer Vercel's head of data science built from a model of revenue broken into its smallest causal units, enriched with first- and third-party signals. Every other agent uses that layer too. "Good data equals good agents."

Build versus buy, every quarter: Vercel's customer support agent costs $300 a month in infrastructure plus token spend, about $150,000 a year, handles thousands of technical cases a week and has three engineers. Vercel tried an off-the-shelf product first, didn't see enough results, and built its own in two months. Building isn't automatic, though. Each quarter Grosser asks the GTM engineering team: "Has another company out there built something better, faster, or cheaper? And if the answer is yes, our agent's going to go."

5 Takeaways from Vercel's Go-to-Market Agents

  • Price the agent, not just the outcome - Under $5,000 a year and a fifth of an engineer replaced a 10-person function's work, at roughly 32x ROI.
  • Document the process before adding AI - Shadow the best person, turn each step into a workflow, then run in shadow mode until the expert has no feedback.
  • Ship developer surfaces - Agents call APIs, MCP servers and webhooks; products without them drop out of agentic stacks.
  • Invest in the semantic layer - d0 and every other Vercel agent depend on a clean, queryable model of the business.
  • Re-run build versus buy - Internal teams can now out-ship vendors in weeks, but Vercel retires its own agents when a vendor does better.

What Vercel's Agent Numbers Mean for Sales and Support Leaders

Grosser's argument is that companies that treat go-to-market as a product, document it, then automate it piece by piece will end up on the right side of the shift to agents. The cost figures make the case concrete: once the process and the data are in place, the agents themselves are cheap. For more on the role behind this, see the AI SDR glossary entry and Databricks on an AI-first sales team.