You're gonna hate the e.g. Southeast Alaska route names, where the plane arrives from one location, unloads some passengers, loads others, then takes off for another destination entirely (specifically Anchorage->Juneau->Petersburg-Wrangell->Ketchikan->Seattle), and yet it's the
same flight number the whole way.
They've competing constraints here: they need all flights to be uniquely identified while in the air, but also all flight identifiers need to be easily expressed over audio. So, even switching to e.g. hexadecimal could be difficult, because "A" can sound like "8". So maybe we use phonetic alphabet for those digits?
Oh but wait, this is to say nothing of the devices actually tasked with transmitting/receiving this info. If the software engineers correctly identified the risk here, then the flight numbers are being stored as two 16-bit integers inside the firmware, so there's plenty of headspace. But if not (say, they only gave 14 bits to the flight number, and 10 bits to the airline designator), then the firmware has to be updated. The FAA lists 648 air traffic control installations in the US, covering 19,000+ American airports. For what are probably obvious reasons, the lions's share of the equipment is not set up for firmware updates via a simple push mechanic, if its setup for remote update at all.
All of this to say, I agree, its dumb, but when this stuff was invented, the prospect of almost 20,000 airports in the US (so, probably on the order of 100,000 globally) likely seemed impossible. Future-proofing has a physical cost, and at the time, it may have been literally impossible to calculate the true cost.