Cursor Cloud Agents Hit 99.9% Uptime as Enterprises Scale Multi-Agent Coding

Cursor, the AI-powered IDE that has become a standard at major enterprises, is pushing multi-agent coding into production with cloud agents that promise 99.9% reliability and instant startup. As of February 2026, companies including Dropbox, Salesforce, and NVIDIA are running hundreds of agents in parallel, with Cursor publishing research on scaling to thousands of concurrent agents and "towards self-driving codebases."

Background

Cursor launched in 2023 and grew largely through organic adoption among developers. The company crossed $1 billion in annualized revenue and reached a $29.3 billion valuation after a $2.3 billion Series D round in late 2025, with backers including NVIDIA and Google. Its differentiation has been an IDE built from the ground up for AI-assisted and agentic workflows, rather than bolting AI onto an existing editor. In 2025 it introduced background agents that can run in the cloud, and in Cursor 2.0 it doubled down on multi-agent coordination and a dedicated coding model called Composer.

Cloud agents let developers send long-running tasks to Cursor's infrastructure so work continues when the developer is offline. The company has improved the experience of sending agents to the cloud from the editor and is rolling out a refreshed UI for managing agents. Enterprise teams can now run agents at scale with the reliability expected of core tooling.

Key Details

Dropbox has emerged as a flagship deployment. The company indexes over 550,000 files with Cursor and accepts more than one million lines of agent-generated code every month. According to Cursor, Dropbox has built an "AI-native" software development lifecycle on top of the platform, with pull request throughput and cycle time in the upper tier of industry benchmarks. More than 90% of Dropbox engineers use AI tools weekly, with Cursor as a primary driver.

Salesforce reports that over 75% of its developers now use Cursor, with some internal measurements showing adoption above 90%. The company has seen double-digit improvements in cycle time, PR velocity, and code quality. One team cut legacy code coverage time by 85%, and unit test generation has increased sharply. NVIDIA has embedded Cursor across its SDLC, with over 30,000 developers using it daily for code generation, testing, debugging, and deployment.

Cursor's Composer model is positioned as a frontier coding model that is roughly four times faster than similarly capable models, favoring interactive flow over raw intelligence. The company has also published research on multi-agent coordination, including a "FastRender" experiment where roughly 2,000 agents ran concurrently and produced over one million lines of Rust for a from-scratch browser project over about a week.

Impact

For engineering leaders, the shift is from "should we try AI coding?" to "how do we run it reliably at scale?" Cloud agents with 99.9% uptime and instant startup reduce the operational risk of depending on long-running agent tasks. Enterprises that have standardized on Cursor are reporting measurable gains in throughput and quality without proportional increases in defects, according to internal and third-party studies.

A University of Chicago study analyzing Cursor usage found that companies merged about 39% more pull requests after the agent became the default mode, with no significant increase in revert rates. Experienced developers tended to accept agent-generated code at higher rates than junior developers, suggesting that effective use of agents depends on clear task specification and context management.

What's Next

Cursor has signaled that cloud agent UI and management will continue to improve, and that multi-agent research will move closer to product. The industry is watching whether other IDEs and platforms can match the combination of in-editor flow, cloud execution, and enterprise adoption that Cursor has achieved. For teams already on Cursor, the focus is on rolling out cloud and multi-agent workflows more broadly and tightening governance and observability around agent-generated code.