Re: Am I missing something?
Yes, you are. Making something that can reliably go to a programmed spot and crash or explode is relatively easy, though not cheap. Think cruise missiles, V-1s, etc. - none of them are cheap unless you have a full metal military budget to play with.
The only 'relatively cheap' small drone I'm aware of that can go some place, do something useful, come back and land would be the Aerosonde,
originally developed as a reusable meteorological drone that cost $10,000 or less. One flew the Atlantic in 1998, but its latest versions look as though they may be a bit bigger and pricier than the 1994 version. These were launched automatically from a car's roofrack and, although they could land automatically in a large flat field, were typically hand flown for landing if the landing area was small and/or had nearby trees etc. to dodge.
Then there was Maynard Hill's 'Spirit of Butts Farm', which weighed 5kg at launch and crossed the Atlantic in 40 hours in 2003, but it was hand-flown for launch and landing. I think he only lost four before one made it across from Newfoundland to Ireland:
The problem is landing: launch and flight to a GPS target are fairly simple to automate, but anything that can land to deliver a parcel and then fly home again will need much better AI than anything yet invented if it is to reliably find a suitable place to land at or near its destination and do so regardless of weather or obstacles and without hitting anybody, their pets or moving vehicles.
And you still need some way of stopping light-fingered gentry from nicking the stuff it is delivering, or the entire drone for that matter.