Cursor’s coding‑agent masterclass

Coding agents are software helpers that can search a codebase, edit files, run terminal commands, and test their own changes. Cursor is now packaging that workflow into a free 30-minute masterclass aimed at developers who want to use those agents more deliberately. (cursor.com) Cursor’s product page says its agent can be started from the desktop editor, command line interface, GitHub, Slack, Linear, and JetBrains tools. The company describes the system as “one agent across every surface,” with the same assistant handling autonomous coding tasks and code editing. (cursor.com) The training lands after Cursor spent the first months of 2026 pushing agents beyond autocomplete. On February 24, 2026, Cursor said its cloud agents could run in isolated virtual machines, test software, and produce artifacts such as videos, screenshots, and logs for review. (cursor.com) On March 5, 2026, Cursor added automations that trigger agents from events such as a Slack message, a GitHub pull request, a Linear issue, or a PagerDuty incident. Cursor said those automated agents can run on schedules, use webhooks, and store memory from past runs. (cursor.com) That is the practical problem this kind of lesson is trying to solve: once an agent can plan, code, test, and react to outside events, the bottleneck shifts to how engineers scope work and review output. Cursor’s January 9, 2026 guide says the “most impactful change” is planning before coding and recommends Plan Mode for tasks that span multiple files. (cursor.com) Cursor describes an “agent harness” as three parts: instructions, tools, and the model itself. Its January guide says the harness matters because different models respond differently to the same prompt, so developers need repeatable ways to set context, choose tools, and check results. (cursor.com) The company is also building more structure around those workflows. Cursor’s Learn site now has separate sections for agents, working with agents, and customizing agents, while its docs include pages for subagents and agent skills that package reusable knowledge and scripts. (cursor.com) Cursor has used its own product metrics to argue the shift is already inside the company. In its February 24 post, Cursor said more than 30% of the pull requests it merges are created by agents operating autonomously in cloud sandboxes. (cursor.com) The masterclass fits that broader push: less as a launch of a new feature than as an attempt to teach developers how to hand off bigger chunks of software work without losing control of the code review step. Cursor’s recent product and training pages all frame the same change in similar terms — delegate implementation, keep the human focused on decisions. (cursor.com)