It’s also true that the ssds are so fast today that the RAM is not as critical, or, better to say, the penalty of not having enough RAM is not as dramatic as it was a couple of years ago.
Yes and no in a way. While SSDs in these Macs offer tremendous performance rivalling the RAM in way older machine - The bandwidth sits between DDR-200 and DDR-400. Well, there's just also more data flying through the chip and everything has grown hungrier. macOS is intelligent about the use of swap and generally avoids swapping an entire program, but instead just bits and pieces so nothing is really slow to deal with but while the SSDs offer bandwidth between 2 and 3GB/s, the LPDDR4X on the SoC is more in the neighbourhood of ≈56GB/s. And that's just bandwidth, way more crucial to the feel of using the device is latency, where DRAM is in the range of a couple hundred CPU cycles and the SSD is probably well over a thousand cycles though I have no exact figure for M1 where the SoC is also acting as the SSD controller.
That said, both Firestorm and Icestorm also have incredibly wide out-of-order re-order queues to keep themselves busy and schedule reads from memory or I/O and performing other tasks while waiting on the read.
The M1 also would use a little less memory for dedicated hardware DMA operations as it has a shared SoC level memory structure and most chips in the system are consolidated into that single system package so you don't need to keep duplicate copies of data from other devices - Though that is negligible on devices that have previously used iGPUs since there's not much other hardware in there with large independent memory space with consolidation pointers to system RAM. It could reduce RAM requirements of devices with dedicated GPUs somewhat though since normally, most data is kept both in GPU VRAM and in system memory and synchronised at certain points when there needs to be cross-hardware communication. Of course now that system RAM also belongs to the GPU but at least there's a smaller need for more RAM chips.
Though even if macOS is less RAM intensive on M1, there's no reason to believe the slew of Electron apps and others will be. Perhaps they will suffer a smaller hit from reaching swap with fast SSDs and the M1 acting as the controller with perhaps more direct access to the NAND, but I doubt the capacity needs are that much less, if at all, before you do hit swap. And swap is still orders of magnitudes slower, even if that speed penalty is hidden by good OOO reordering.
All in all my personal conclusion is thus:
Look at your workload in Activity Monitor on your current Mac. How much RAM do you have and what colour is the pressure graph? Make your decision based on that.
8 could very well be enough, but you can't change it later, so if you're pushing into the yellow or red with 8GB currently, I would get 16 for the M1 just to not regret it even if it's not felt as hard. And it while it might cost more I also do think the resale value for the future will be slightly higher; Not entirely offsetting the difference, but minimising it somewhat.
Don't buy excessively, but don't expect it to handle memory exactly like iOS - The macOS and iOS memory management systems and how much extra macOS needs to deal with is still a difference between the two platforms and macOS' multi-tasking model does not have the same suspend-stages iOS does. Apps in the background run more actively (iOS apps on the Mac somewhat being in a funky middle stage here since the code paths have the suspend stages and the associated frameworks are there, but the Mac will prefer having them more active even when backgrounded so that they respond immediately if things happen and can give visual cues even when they're not in the foreground, though it can still deep-sleep some aspects of the execution - WWDC talk exists that goes into more detail on that)
