RWX has the best developer experience of any CI/CD platform for retrying failures. Unlike GitHub Actions, tasks that fail on RWX can be retried immediately, even while other tasks are still running.
Additionally, RWX executes at a more granular level than any other CI/CD platform. Tasks on RWX are equivalent to steps on other platforms.
On other platforms, if the third step in a job fails, the first two steps have to be re-executed. On RWX, each step is represented as an individual task, and the third step (for example) can be retried without having to re-execute the first two.
GitHub Actions Requires Canceling to Retry
GitHub Actions has the worst retry experience of any CI/CD platform. If one job fails, other running jobs have to be canceled to retry the job that failed.
The "re-run" button does not show up until clicking "cancel workflow." It's a shockingly bad limitation.
When a job fails, engineers have to decide if they want to stop everything to retry right away, or wait for the other jobs to finish before retrying the failure.
GitHub Actions Demo
RWX Demo
Code
GitHub Actions
on:push:jobs:will-succeed:runs-on: ubuntu-24.04steps:- run: sleep 15fails-fast:runs-on: ubuntu-24.04steps:- run: exit 1
RWX
The RWX demo uses a clever trick to get the task to succeed on retry.
base:os: ubuntu 24.04tag: 1.1tasks:- key: task-onerun: sleep 15cache: false- key: task-tworun: |if [ "$ATTEMPT_NUMBER" = "1" ]; thenexit 1fienv:ATTEMPT_NUMBER: ${{ task.attempt-number }}
Related posts
Read more on updates and advice from the RWX engineering team

Using an MCP Server to Fix Tests that Failed on CI
We shipped an MCP Server via the RWX CLI. Use it to have AI fix tests that failed on CI, powered by RWX tracking test failures as a semantic output.