Two software companies have each published accounts of agents they built for their own revenue teams, nine months apart. Vercel says its lead qualification agent helps one SDR do the work of 10. Salesforce says a three-person squad shipped a seller work-list app in days, and the app has generated 28,000 opportunities.
What Vercel reported
In a post dated February 9, 2026, Vercel wrote about building and running agents on its own platform. It lists its lead qualification agent among its examples and says the agent “helps one SDR do the work of 10”. The post links a template for the agent and gives no further figures for it, no baseline and no date for when the agent went live.
The post gives more detail on a second internal tool, a data agent called d0. Vercel says d0 is a text-to-SQL engine, that one person built it in a few weeks using 20% of their time, and that it answers hundreds of questions a day. The argument of the post is that companies of any size can now create agents quickly, and that the work left over is running them reliably, with attention to security, observability, reliability and cost.
Advertisement
What Salesforce reported
On November 4, Joe Inzerillo, president of enterprise and AI technology at Salesforce, described how Salesforce builds internal agents. The method is called Rapid Innovation Squads. Each squad starts with a business stakeholder who owns a problem and has strong views on how to solve it, adds between one and five agentic builders, and shrinks the problem to its smallest useful form. The squad records its morning stand-up, posts the transcript to a public Slack channel and hands it to an agentic coding tool, which keeps building and testing overnight.
The post gives three results from that method.
- A three-person squad built the Sales Signals Slack app for small-business sellers, which gathers website visits, usage spikes and renewal windows into a daily work list. Salesforce says the app has generated 28,000 opportunities.
- A three-person squad built a Slack skill that pre-fills solution engineers’ timesheets from their calendar, Slack data and Salesforce records. Salesforce says it took less than four weeks, adoption is at 80%, and 23,000 weekly activities are now logged automatically.
- A squad in the legal department combined deterministic rules with Agentforce to route cases that used to take about 20 minutes each across roughly 39,000 cases a year. Salesforce says it saves $350,000 a year.
The post also says 74% of Salesforce employees have used an agentic coding tool such as Claude Code or Codex at least once, and that usage among nontechnical employees is up to 66%, more than triple the first quarter. Salesforce plans a company-wide version in December called RISFest.
What the two accounts share
Both describe a narrow workflow with a named owner and a countable output: leads qualified, opportunities created, activities logged. Salesforce’s squads each started from one problem shrunk to its smallest useful form, and Vercel’s post links a template for its lead agent.
Both also come from the companies themselves, and no outside party has checked the figures. The Salesforce post does not say over what period the 28,000 opportunities were counted or what the baseline was, and Vercel’s post gives the 10-to-1 ratio without a method. Treat them as reported results. For wider market context, see our coverage of Menlo Ventures’ enterprise AI report, which found limited evidence of AI replacing CRM.
Where each says the guardrails sit
Salesforce says IT still owns the systems of record and the data semantics, such as what a customer is or what revenue means, and exposes them through APIs and MCP servers. Vercel puts the difficulty in operations. Its post says production deployment takes expertise in security, observability, reliability and cost optimization, and that those skills remain rare.
The two positions line up. A seller-facing agent reads CRM fields and writes back to them, so the definitions of those fields decide whether the agent is right. We made a related argument in an opinion piece on adding an author-type field to the CRM before agents write at volume.
What it means for the sales leader
In our reading, these accounts give a sales leader a template for a first internal agent, with the figures still to be proven. The template has four parts: one workflow, one owner who knows the work, a unit that can be counted, and a baseline recorded before the build starts. Qualifying inbound leads, logging activity and compiling a daily seller work list all fit that shape.
Ask RevOps for the current numbers on the chosen workflow before anyone writes a prompt: how many leads arrive, how long each takes to qualify, how many hours sellers spend logging. Those are the figures Vercel and Salesforce left out of their posts, and they decide whether a result like 10-to-1 holds in another company’s pipeline.
Vendor evaluations change too. A team that can build a narrow agent in weeks has a point of comparison when a vendor quotes a seat price, which connects to our earlier opinion that sales leaders should ask agent vendors to price on work done.
Inzerillo closes his post with advice that applies to any of these projects. “Don’t stop at automating a task. Redesign the whole workflow so every step runs at agentic scale,” he writes. His first step is to “Start with a problem and an opinionated stakeholder who owns it.”
Source: Salesforce News
