The idea
Continuous integration runs checks on every change: linting, type checks, unit tests, evals. Continuous delivery or deployment builds an artefact (a container image) and ships it to staging and production automatically when checks pass.
A typical AI pipeline: on pull request, run tests and the CI eval subset; on merge to main, build and push an image tagged with the commit SHA, deploy to staging, run smoke tests, then deploy to production (automatically or after approval).
Pipelines make releases boring and repeatable, and every running version traceable to a commit.
Manual deploys are slow and error-prone. CI/CD is how teams ship many times a day safely.
Picture it
Words to know
- ArtefactThe built output that gets deployed, like an image.
- Smoke testA quick check that the deployed app basically works.
Try it without code
Map your pipeline
- Write every step you currently do by hand to ship your app.
- Mark which ones a machine could do.
Use it in code
The skeleton of a CI/CD workflow (details in the next lessons).
name: ci-cd
on:
pull_request:
push: { branches: [main] }
jobs:
test:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: actions/setup-python@v5
with: { python-version: "3.12" }
- run: pip install -r requirements.txt -r requirements-dev.txt
- run: ruff check . && pytest -q
deploy:
if: github.ref == 'refs/heads/main'
needs: test
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- run: echo "build image, push, deploy to staging, smoke test, deploy prod"Write down your pipeline stages and add the test job to your repo today.
Common mistakes
- Pipelines nobody trusts because tests are flaky.
- Deploying from laptops "just this once".