With the rise of agent coding, the working life of software engineers has become incredibly complex. A single engineer may oversee dozens of coding agents at once, initiating and directing various processes as needed.
There was so much to keep track of, and the attention spans of human engineers quickly became resource limiting.
Cursor on Thursday released a new tool aimed at curbing that confusion. This new system, called Automation, gives users a way to automatically launch agents within their coding environment, triggered by new additions to the codebase, Slack messages, or simple timers. As Cursor explains, this is a way to review and maintain all new code created by agent tools without having to track dozens of agents at once.
At the most basic level, automation is a way for engineers to break out of the “prompt and watch” dynamics that define most agent-based engineering. Instead of starting an agent at a human prompt, Cursor’s automation framework allows you to start the agent automatically and loop in the human whenever you need it.
“Humans are not completely ignored,” Jonas Nelle, Cursor’s head of engineering for asynchronous agents, told TechCrunch in an interview. “It’s that they don’t always have the initiative. They’re called at the right point on this conveyor belt.”
One of the early examples is Bugbot. This is a long-standing cursor feature that the team sees as a precursor to a broader automation system. The Bugbot system is triggered every time an engineer makes an addition to the codebase and reviews the new code for bugs or other issues. Using automation, Cursor was able to scale its system to more complex security audits and more thorough reviews.
“This idea of thinking harder and spending more tokens to find harder problems was really valuable,” said engineering lead Josh Marr.
tech crunch event
San Francisco, California
|
October 13-15, 2026
Cursor estimates that it runs hundreds of automations per hour, far beyond simple code reviews. This system is also used for incident response, and when a PagerDuty incident occurs, it starts an agent that can instantly query server logs through an MCP connection. Another automation provides a weekly summary of changes to Cursor’s company codebase on Slack.
“Abstractly speaking, anything that automation starts can also be started by humans,” Nelle says. “But automating this changes the kinds of tasks that models can perform within your codebase.”
This new system comes amid intense competition in the agent coding space, with both OpenAI and Anthropic making significant updates to their agent coding tools in the last month.
Ramp data shows that Cursor’s market share has remained stable since May, with approximately 25% of generative AI clients subscribed to Cursor in some way.
Still, the company’s revenue continues to grow at an impressive pace due to overall growth in the agent coding space. Earlier this week, Bloomberg reported that Cursor’s annual revenue has grown to more than $2 billion, doubling in the past three months.
