Cursor Enterprise Results: NVIDIA, Salesforce, Stripe

Cursor's enterprise page reports its reach in the figures below. They describe reach and preference, and Cursor reports them itself, so they are not independently audited counts.

The last two need the most care in front of a buyer. The page doesn't describe the head-to-head evaluations behind the 93%. The 100M+ lines count code written with Cursor, and the page doesn't say how much of it an agent wrote. Neither number can be reproduced from outside.

Cursor's enterprise page quotes NVIDIA's president and CEO, Jensen Huang, by name.

The quote continues: "Every one of our engineers, some 40,000, are now assisted by AI and our productivity has gone up incredibly." Cursor's NVIDIA customer story, from February 2026, adds measured figures. Over 30,000 developers use Cursor daily, they commit three times more code than before, and bug rates have stayed flat while code style has grown more consistent. The two counts measure different things: Huang's 40,000 are engineers assisted by AI, and the story's 30,000 are daily Cursor users.

The story says NVIDIA's customizations took the impact beyond individual productivity. One team uses custom rules to automate its git flow, from branch creation and commits to debugging continuous-integration failures and tracking issues. Another team's bug-fix automation pulls context from tickets and documentation through MCPModel Context Protocol. A standard that lets an AI agent pull in context from outside the repo, like Jira tickets or internal docs. Press Enter for the full definition. servers, then has Cursor implement the fix and run tests. In a Cursor talk, a presenter also said NVIDIA uses hooks and other enterprise controls to keep Cursor within its compliance boundaries, which the story doesn't mention.

Cursor's Salesforce customer story reports that over 75% of Salesforce developers use Cursor and that PR velocityHow quickly pull requests are merged; the easiest delivery metric to measure, though it sits furthest from the customer outcome. Press Enter for the full definition. rose by more than 30%. Salesforce tracks three metrics on dashboards it already had, which made the change visible quickly.

A Cursor talk on developer productivity gave different Salesforce figures. It said more than 90% of Salesforce's engineers use Cursor, across about 20,000 developers and a monorepo near 30 GB, with roughly 2x gains in both code quality and velocity. The talk said Cursor's codebase indexing and large-codebase understanding improved cycle time, quality and throughput. The written story repeats none of those numbers.

Cursor's Stripe customer story, from February 2026, describes how Stripe got more than 3,000 engineers started on Cursor within guardrails. The story says Stripe is known for 99.9999% API uptime and for products that set the bar for developer experience, and could not sacrifice either.

• Cursor comes preinstalled on every developer's machine, and new hires learn the Stripe development environment in a lab.
• Cursor rules give the agent context about Stripe's codebase and coding standards, and teams can add their own.
• Code review changed too: AI flags complex methods and risky files, so reviewers know where to look.
• On an internal survey, developer sentiment hit a five-year high.

One finding surprised Stripe's head of developer infrastructure. He expected junior engineers to gain the most. Instead, he has seen the longest-tenured engineers make the biggest productivity gains. His explanation is that they already have the context in their heads, so they can see what they want and direct agents to it.

The same Cursor talk on developer productivity described Stripe results that the story doesn't state: less setup and debugging time, and prototypes in hours rather than days. It quoted Stripe's head of developer infrastructure saying AI and Cursor "opened up new skill sets". The talk's examples were a front-end engineer contributing to back-end work and an iOS engineer exploring another part of the stack.

Another Stripe data point runs the other way. In Cursor's workshop for finance and accounting teams, a presenter said Cursor's finance team used an agent to build a detailed IC+ processing-cost analysis, then sent it to Stripe to confirm the findings.

The presenter added that the analysis found real categories where Cursor should have been optimizing to save money. Treat it as an internal-team result with outside sign-off, not an audited Stripe metric.

Cursor's July 30, 2026 post on its cloud agent environment says cloud agents write more than half of the PRs merged to Cursor's monorepo, up from roughly one in ten the previous December. In February 2026, Cursor's post on the third era of AI software development had put the share of PRs it merges internally at 35%.

A Cursor talk put the share at almost 40% and tied the jump to Cloud Agent Artifacts: after they shipped, the cloud-agent share of merged PRs roughly tripled. The same talk said that by about March, roughly 99% of Cursor's generated code came from cloud or IDE agents. Neither written post gives these talk figures. All of this is Cursor reporting on Cursor, and the share changed within months, so date any figure you quote.

Some results come from people who don't write product code. In Cursor's workshops, presenters described sales engineers, go-to-market teams and designers who build their own tools instead of filing a ticket and waiting for an engineer. These are anecdotes from single workshops, not measured results like the NVIDIA or Salesforce figures.

Airtable: from no-code demos to bespoke coded ones

In Cursor's workshop for sales and field engineers, a presenter described Airtable's sales engineers. They built customer demos in a no-code app builder with about a dozen preset components, plus "interface extensions" that allowed custom code, and that is where Cursor came in. Once every Airtable SE had a Cursor license, demos moved from preset templates to bespoke builds that wrote TypeScript into the demo environment, for uses such as financial reporting and dashboards.

According to the presenter, a custom demo used to mean asking an engineer and waiting a week or two. With Cursor, sales engineers who could vibe-code built custom demos as part of their normal work, so more calls could include one.

Two workshop stories come from inside Cursor: a sales tool built by someone who isn't an engineer, and a prototype copy of Cursor that its design team uses to try ideas.

ChatGTM: an internal app nobody set out to build

In Cursor's knowledge-work workshop, George, a self-described non-engineer on Cursor's go-to-market team, described a common prospecting problem. A rep can spend two to three hours researching one account across five tabs: the CRM, call recordings, usage dashboards, LinkedIn and enrichment data. George didn't set out to build an app. He solved one problem at a time inside Cursor.

1. 1Describe in plain English what data exists and where it lives. The agent writes SQL to consolidate it, and reps check the output ("these numbers look too low" or "looks good") and iterate.
2. 2Describe the scoring logic in plain language: seniority, how recently the account used the product, when the lead came in. The agent writes a Python scoring file, which the team tests against experienced reps' judgment and tunes.
3. 3Write outreach guidelines, not rigid templates, so the agent researches each account and drafts a message that fits it.

Demand grew from there: pulling contacts, choosing whom to contact, drafting messages, building account plans and preparing renewals. The team wrapped the same tools into an internal app called ChatGTM, built in weeks rather than months, per the workshop.

Prototyping with "baby Cursor"

Cursor's design team built "baby Cursor", a standalone instance of Cursor, according to the presenter in Cursor's workshop for product managers. It is simpler on the backend but full-featured in the user experience, and the team uses it only to iterate on prototypes and try new interaction patterns. An idea can then be shown as a working prototype instead of described in a PRDProduct Requirements Document. The spec describing what to build, for whom and why, before implementation starts. Increasingly the artifact you hand an AI agent as the source of truth for a feature. Press Enter for the full definition..

Every figure on this page comes from Cursor or from the customer it describes, and none comes with an independent audit. That doesn't make them false, but it puts the work of context on you. Before a number goes into a deal deck, check its date, its source and whether it describes one company.

When a Cursor talk and a written customer story give different numbers for the same customer, as they do for Salesforce, quote the written story. It carries a date, and anyone can check it.