I'm a bit trepidatious weighing in to this thread, but here goes anyway! As I work in front of a computer all day long I never thought that I'd get interested in CNC, but 20 years or so ago I started to put a workshop together and got involved in "model engineering" (though I've never built a model and don't suppose I will). Initially I was getting fed up with changing change wheels on the lathe and never having a fine enough feed so started to investigate "digital screwcutting". The lightbulb moment was when Tony Jeffree published his ML7 CNC conversion and I realised that it would be just as easy to add a second stepper to the lathe and just use a general purpose driver like Mach3 rather than a single-use package. Since then I've added a CNC mill (not converted by big VMB yet) and digital dividing.
I really don't recognise what some people are saying about CNC in this thread. In my experience….
It isn't about "programming". I seldom or never directly write G-code. I use the wizards in Mach3 Turn a lot, but I must say that you need to keep a careful idea on their output otherwise the tool can do some unnerving and dramatic moves you don't expect! If I have to turn to a shoulder, that takes about 2 minutes setting up the wizard which generates the code and the machine runs. But setting the tool accurately, selecting the right feed rate and DOC, are key to getting the result you want, and you need to use the wizards as just a starting point if you really want accuracy. This takes skill and experience – just as much IMHO as manual turning, and at least you don't have to stand in front of the lathe inhaling coolant fumes while you twiddle handwheels. As for threading, there is no comparison – just look at the endless debate on change wheels on this forum, and "can you cut metric threads on an imperial lathe?". With CNC, it's a doddle, any (sensible) pitch, at normal turning speed, up to a shoulder, the machine does the pickup.
"CNC is only for quantity production" – rubbish. I have just started making Woodward's Gearless Clock, with the count wheel (which looked to me the hardest bit). I could have crossed it out with a piercing saw and filed to the line once I'd cut the teeth. But I would probably have made a mistake and might have had to start again. But I spent an enjoyably challenging hour or so writing a spreadsheet to work out the dimensions for any 4-spoked wheel cross-outs and directly generate the Gcode. I made a simple jig to hold the wheel, held it down with a bolt and some shellac, and did the crossing out on my little CNC mill. A job I found full of interest, and much more appealing than hunched over a cutting-V breaking saw blades and getting brass dust in my socks. There will be another interesting challenge in producing the 'scape wheel, and I'm especially looking forward to the design of the daisy wheel! When I've finished I'll have some jigs and programs to make more clocks if I want…
A while back I looked at producing gear cutters using the "button" method, and posted on here describing how I did it (the practical bit at least). Again that involved writing a spreadsheet that started with Mike Cox's formulae and generated the Gcode to drive the tool to turn the profile on the cutter. With this I don't have to make different buttons for each module, I can make cutters for modules and pressure angles that are impractical by the conventional method, and this would have not been possible without CNC! Similarly I've generated code for tapers that can't be made with the Mach 3 wizards. I have a plan for a pendulum compensation scheme that needs a carefully shaped cam – not possible manually, trivial with CNC.
How many plans have been published for ball turning tools? A hardy perennial. Redundant with CNC. You want to turn an ellipsoid? a hyperboloid? Easy.
Having a CNC mill, I was easily able to knock out an engraved plaque for a Jubilee Oak Tree in our village. And to produce a small memorial plaque for my woodworker father-in-law's ashes engaved on a sliver of oak from the Victory (don't ask how he got it).