Engineering2 min read
Shipping weekly without breaking things
Small pull requests, feature flags and a deliberately boring CI pipeline let us release every week — even on products with thousands of daily users.
Most teams don't ship slowly because they lack talent. They ship slowly because every release feels risky, so releases get batched, and batched releases are risky. Breaking that loop is mostly about habits, not tools.
Here's the setup we bring to every product we build.
Keep changes small
A pull request should be reviewable in about fifteen minutes. If it isn't, it's two pull requests. Small changes are easier to review properly, easier to test, and — when something does go wrong — easy to revert without dragging three unrelated features down with them.
We aim for most pull requests to stay under 300 changed lines, excluding generated files.
Hide unfinished work behind flags
Trunk-based development only works if half-built features can live in production safely. Feature flags make that possible: code merges early, but users only see it when it's ready.
export function canUseNewCheckout(user: User): boolean {
// Internal team first, then a 10% rollout, then everyone.
if (user.isStaff) return true
return flags.isEnabled('new-checkout', { userId: user.id, rollout: 10 })
}
Flags also give you an instant off switch, which is worth more than any rollback script.
Make the pipeline the gatekeeper
Reviews catch design problems. Pipelines should catch everything else, automatically and every time:
- Type checking and linting
- Unit and integration tests
- A production build
- Preview deployments for every pull request
- Smoke tests against staging after each merge
If a check is flaky, we fix it or delete it. A pipeline people learn to ignore is worse than no pipeline.
Release on a schedule, not on a feeling
We release at least once a week, on the same day, with release notes generated from merged pull requests. Because each release is small, the notes are short — and if a release needs a "big launch" conversation, that's a sign the work wasn't split finely enough.
The goal isn't speed for its own sake. It's making each release so small that shipping stops being an event.
Watch what happens after
Every release is followed by a short look at error rates, latency and the key product metrics. Monitoring isn't a launch-week task; it's part of the definition of done.
None of this is exotic. That's the point: boring, repeatable delivery leaves your team's energy for the parts of the product that actually deserve it.
About the author
Theo Park · Co-founder & CTO
Theo leads engineering and architecture. He cares about fast feedback loops, boring deployments and code that is easy to change.