Cursor builds always-on agents to tackle developer task tedium

AI-powered code editor company Cursor has introduced Cursor Automations to build always-on agents that operate on its deep understanding of the codebase.

Established as a fork of Microsoft’s extensible VS Code code editor, Cursor is known for its Plan Mode, which helps developers create detailed implementation plans before writing any code; its Composer interface, which allows users to build features based on natural-language instructions; and its Tab predictive autocomplete functionality.

As agents themselves have become increasingly smart, the Cursor team says that they found themselves “kicking them off for the same kind of task over and over again,” so, logically, they thought, why not automate that process too?

Dead code cleanup

Example use cases include an incident triage process that run everytime a PagerDuty monitor goes off, or a personal automation that looks at all a developer’s pull requests on a given day and then works to clean up dead code or bad programming patterns (such as deeply nested conditionals where an excess of if-then statements hide core system logic) to make developers’ lives easier.

Cursor’s engineering team member, Jack Pertschuk, and Cursor software engineer, Jon Kaplan, explain more of the rationale behind Cursor’s genesis and its latest augmentations in a video released on March 5.

“All those things that people are used to configuring with checkboxes don’t have to be done anymore.”

“The way we have changed software application development has changed so much in the last nine months — and with more software and more output, there’s more stuff that you need to review, more issues you need to triage and more things to manage around software, but a lot of these ‘things’ are automatable. All those things that people are used to configuring with checkboxes don’t have to be done anymore… just prompt your agent,” say Pertschuk and Kaplan.

Events trigger agents

These agents run on schedules or are triggered by events like a sent Slack message, a newly created issue in Linear (the high-performance, keyboard-centric project management tool), a merged GitHub pull request, or (as noted above) a PagerDuty incident. In addition to these built-in integrations, developers can configure custom events using webhooks.

Once invoked, an automated agent spins up into life in a cloud sandbox. It then follows the developer’s instructions using the Model Context Protocol connections and models configured by the developer, and then verifies its own output. Agents also have access to a memory tool function that enables them to learn from past runs and improve with repetition.

Users appear to like Cursor. Tim Fall, senior staff software engineer at workforce management platform company Rippling, says in a Cursor blog post, “Automations have made the repetitive aspects of my work easy to offload. By automating task rounding, handling doc updates, and responding to Slack messages, I can focus on what matters. Anything can be an automation!”

Nifty on nits

The Cursor team explains how it views automation as eminently useful for tasks such as reviewing code changes. Automations in Cursor are able to catch and fix everything from style nits (code formatting preferences that don’t necessarily initially affect application functionality) and other inconsistencies to security vulnerabilities and performance regressions.

In the video, Pertschuk and Kaplan refer to Cursor Bugbot (first launched in June 2025 as part of Cursor 1.0, the tool that reviews pull requests and identifies bugs, security issues, and code quality problems) as the “original automation” that runs on every pull request.

Now that the platform as a whole is more evolved, the pair says that automations allow users to customize all kinds of review agents for different purposes.

Dogfooding dinners

By way of eating its own dogfood, the Cursor team writes in the announcement blog post, “Our security review automation is triggered on every push to main. This way, the agent can spend more time finding more nuanced issues without blocking the PR. It audits the diff for security vulnerabilities, skips issues already discussed in the PR, and posts high-risk findings to Slack. This automation has caught multiple vulnerabilities and critical bugs at Cursor.”

In terms of regularity, Cursor starts every morning with bacon and eggs, then sends an automated agent to review recently merged code. That agent identifies areas that need test coverage. It follows existing conventions when adding tests and only alters production behavior when necessary. The agent then runs relevant test targets before opening a pull request.

On a weekly basis, Cursor delivers an automated Slack digest summarizing meaningful changes to the repository over the last seven days. The agent highlights major merged PRs, bug fixes, technical debt, and security or dependency updates.

Developer Death Star?

Are toolsets like Cursors part of the recipe that will spell the end of software development itself? Obviously, it’s quite the opposite.

Jensen Huang, president & CEO of Nvidia, says in a statement on the Cursor website that his “favorite enterprise AI service” is Cursor. He further states that every one of his company’s software engineers, some 40,000, is now assisted by AI, and their productivity has gone up incredibly.

As we stand in 2026, we are perhaps now seeing the full-color version of what automation tools can do at the command line and how they impact and improve software engineering workflow practices. There is still much to learn and evaluate in this space, but there may be more meat on the bones now.