Both hit a running system, so they get put in the same suite and run at the same time. They answer different questions and should fail at different points in the pipeline.
# smoke: is this deployment alive? seconds, runs after deploy
curl -fsS https://shop.example.com/health
curl -fsS https://shop.example.com/ | grep -q '<title>'
# integration: does checkout work against a real database?
# minutes, runs before deployA smoke test is a handful of requests proving the release is serving traffic, and it is the gate on whether to roll back — so it must be fast enough to run immediately after a deploy and stable enough that a failure means something. An integration test exercises behaviour and belongs before the deploy. Mixing them produces a slow gate that people learn to ignore.