Usually by making a PCB, yes. Often with toner transfer to make at least the initial prototype, even if it's not adequate for the final circuit board. Often adding extra headers and pads and test points, if confidence not so high. I try to add a spot to solder header pins for anything I may want to scope.
Also I occasionally point-to-point smaller circuits and/or pieces of circuits on protoboard. Regular veroboard works for 8 pins or less SOIC, anyway.
I occasionally also use some 50 mil, hole-less veroboard for SOIC. Homemade. Must solder under a microscope to see anything. Very simple to make with toner transfer and pcb software, even if you don't yet know how to use the software very well. DIP parts can also go on this board, trimming down the legs and soldering to every other pad or even just blobbing to two rows of pads per pin.
But maybe the reason I can go straight to PCB most of the time, because I'm not often creating interesting circuitry. Mostly simple stuff that any hobbiest already knows and which is well documented. For instance, even building "my own" first DC boost circuit was essentially copying the recommended layout from an IC datasheet, lol.
Build first, fix it later, may not work so well if you are making intricate circuitry with discretes and analog. For what I do, mostly I can figure out the smaller pieces I am not familiar with on a breadboard with just a few components. I have the feeling that MOST of the really interesting discrete circuits are already on an IC with cut-n-paste instruction book/datasheet. Maybe I live in a very boring bubble.
But no matter what, it still seems like overall a waste of time to do anything more than a few components on a protoboard or breadboard. Debugging is much easier with schematic AND a clean pcb layout. Even though it takes time to transfer and etch the board, initially (a couple hours), it can usually be easily reworked until solution is achieved. Especially easy with no soldermask. 90% of your initial pcb work is going to finished with only minor tweaking. And I was going to do that, anyway. Especially with a microcontroller, it can be very much a waste of time to breadboard it, making scores of arbitrary pin assignments, then finding out an actual pcb layout is completely retarded. Keeping track of pin changes is a headache. And changing things can create other unintended consequences/bugs. I much prefer to work on the pcb layout, first. Figure out the easiest layout to get ANY pin where needed, then refer to datasheet to make sure it is a suitable pin. It's much easier to make these arbitrary decisions when it's not arbitrary. And it's much better if you can makes these decisions only once. In software. When you click on a real life jumper wire, the name of the signal/pin doesn't automatically pop up. Yes, mistakes happen, and board revisions commonly are needed. It's common because it's still the easiest way to do it.
Often is the case, I may even do some breadboard work AFTER making the pcb. For instance, I often can verify some basic or cricital functionality of my firmware on a breadboard before firing up the pcb to avoid frying things. My DIY 8 channel logic probe comes in handy here, providing high impedance visual indicators, selectable pullup/downs, selectable tactile switch inputs to either rail, plug in for thruput to 8 channel logic analyzer, and place to plug in pin headers where I can hook a scope probe on each of 8 signals.