Oleg has a good solution. I have done this a few times for prototyping. Especially in the days before ISP. The CPU contains the bootloader, monitor, debugger... The NVRAM (or EEPROM) is wired as Oleg described, with !RD and !PSEN ORed to the !RD pin on the NVRAM, and A15 inverted and wired to !CS.

Another trick is to use jumpers to select either
!EA = 1 and A15 ---|>0--- !CS (as described)

or

!EA = 0 and A15 = !CS (not inverted)

You can then use the CPU to load some code to the NVRAM, remove power, move the jumpers, and then the code you just loaded is live at address 0 when you power up.

This works best with code having only AJMPs and ACALLs. If you have to use Long jumps and calls, I compile (or assemble) at address 0, then use a hex file utility (or just a device programmer software) to add 8000H offset. The CPU loads the file to 8000H, but it is identical to if it was loaded to address 0 after the jumpers are moved.