Do you care to have someone (any competent engineer) to fix it when it breaks?

yes: you want an UI as a consistent entry point for workflow diagnostic.

no: it should not be built in the first place.

Is the workflow expected to be fast (millisecond/second)?

yes: you can't afford the overhead of an orchestrator.

no: the added overhead of orchestration is acceptable.

Does the workflow involve different access/connection?

yes: you should maintain minimal access at task level and use the orchestrator to assign access to task when possible.

no: you should still have an orchestrator since individual access is difficult to config/maintain, and you don't want people waiting when they are blocked by access control.

Does the workflow has anything to do with time?

yes: you need airflow.

no: maybe.

Does the workflow involve task dependency?

yes: you need airflow.

no: you don't need airflow.

Do you like to reinvent the wheel?

yes: sure.

no: use airflow or any other workflow management tool.

Do you like free metrics/dashboard to monitor the workflow?

yes: you like airflow.

no: when the time comes, you might regret it.

Do you want to have auto retry for certain task/workflow?

yes: you like airflow.

no: everything breaks for random reason but might the issue not persist, retry makes your life easier.

Does airflow increase the risk of workflow failure due to failure of itself?

Yes, but you should have more gain from above than the added risk.