In the example in your post, the queue is abstracted away in some sense, but you aren't just "writing code". There's step.run(), step.sleep(), step.sleepUntil() sprinkled throughout. I'd say in something like Celery, which is explicitly about running jobs asynchronously backed by a message broker/queue, you really do just write code, but then call the function with .s().apply_async() and so on. Now, I'm not saying you're wrong that we ought to abstract queues away, just pointing out that the abstraction leaks in your workflow example. Happy to hear where/how I'm wrong on this.