Jules

Jules occupies a different slot than Cursor or Claude Code: it's not a pair programmer, it's a contractor. You hand it a task — fix this bug, bump these dependencies, add tests for that module — and it spins up a cloud VM, clones your repo, drafts a plan for your approval, executes unattended and comes back with a GitHub pull request. That async loop is precisely wrong for exploratory hacking and precisely right for the long tail of maintenance work every team defers. Since leaving beta it has matured fast: Gemini 3.1 Pro under the hood, a cross-platform CLI and open API for scripting it into CI, parallel tasks with environment snapshots, and Memory that retains your repo's conventions between runs. The free tier — 15 tasks a day, 3 concurrent — is the most generous in the category and makes trying it a no-brainer. Constraints to know: it's GitHub-centric, so other forges fit awkwardly; there's no standalone paid tier, so scaling means a $19.99 Google AI Pro subscription; and complex PRs still demand real review before merge. Used for what it's built for — routine, well-scoped, parallelizable work — Jules is the most practical async agent you can adopt today, and the price of entry is zero.