That sounds like the best of both worlds. There's always going to be an activation cost to pay, even for the most motivated.
So if I'm inspired, it'd still be helpful to put hands on keyboard and start, otherwise that inspiration might be squandered.
On the other hand, if I'm not inspired and don't become inspired in the first X minutes, there was probably a better chance of becoming inspired because I did start, but also a ceiling on X that indicates it just isn't happening today.
So maximal productiveness = {start regularly, to take advantage of inspiration} + {start regularly, to see if inspiration strikes}
And also worth differentiating forward progress tasks from epiphany tasks. Sometimes a great book still needs 2 more bad chapters to be complete, but a bad modification on a mediocre ML model doesn't help it score better.