Using Cloud Agents

00:00 Cloud agents generally offer the most autonomy of the four workflow types. You describe a task, the agent works in a remote or managed environment, and later the agent reports back with a branch, pull request, or prototype.

00:15 There’s no step-by-step approval, the agent just goes. That makes cloud agents a good fit for greenfield prototyping or work that takes longer than you’d want to sit and supervise.

00:28 Tools in this category include Devin, Claude Code on the web, Codex web, and Cursor’s Cloud Agents. You can often access them through Slack, issue trackers, or a browser.

00:40 For example, you can mention Claude with @claude in Slack, ask it to complete a task, and it’ll spin up a Claude Code session on the web. There’s a trade-off though.

00:52 More autonomy means less real-time control. Your code runs on infrastructure outside of your local machine. That makes cloud agents most useful when the task is clearly scoped and the output is easy to review afterward, like a branch, pull request, or prototype you can inspect before it goes anywhere.

01:13 This is what Claude on the web looks like, and it has access to the repository you created in the previous example. Remember the PR that GitHub Copilot code review left a few comments on?

01:25 Well, now it’s merged in main and the bug has been fixed. However, it is far from ready to go into production. So you could ask Claude on the web to prepare this project for release.

01:37 Add tests for the core functionality, improve the README, and create a pull request. Then you can do something else and in a few minutes it will create a PR for you to review.

01:49 So Claude Code on the web created this PR preparing the project for release, just as instructed.

01:58 Here you can see the summary

02:01 and there’s only one commit.

02:05 It added some tests, added a .gitignore, and refactored some of the code that was already there. Now it’s up to the developer to decide whether this is ready to merge or not.

02:18 And one final note about this type of agent. Not every cloud agent uses the same execution model. Claude Code on the web runs on Anthropic-managed infrastructure.

02:29 GitHub’s Copilot cloud agent runs in an ephemeral environment powered by GitHub Actions. Cursor’s Cloud Agent can also run on machines you control. Find out where your code actually executes before connecting proprietary work to a cloud agent.

02:44 And as with all AI-generated code, review everything before you push. In the next lesson, you’ll see how all four workflow types blur together in the real world.