If somebody is very smart and contributing a lot, then it is fun. If they don't match
that kind of level of energy, then it is really not the right place for them. It is an
exciting thing. It is still a little bit different. People can't come and talk to me
everyday[laughs]. And so they have to look to their Business Unit Manager, which is
how we have it setup. Certainly, we are trying to preserve all of that culture, and get
the advantages of being a large company with a broad product line, with stability,
worldwide presence, great support, and yet have the advantages that a small software
company has.
DKA: Bill, with the AT came a new chip that brought a lot of traditional speed and
capability. How did that affect your business at Microsoft?
BG: After the PC came out in 1981, the next step that IBM took was the XT
where they put a hard disk in. So, we did a version of MS-DOS with a hierarchical
tree oriented file system. The next step after that was the AT 286-based machine. The
286 was quite a bit faster than the 8088 that had come before. Almost four times as
fast as the original 8088 machine. The problem was that if you wanted the 286 to
address more memory, it had to run in a mode that was completely incompatible
with the original chip. And none of the software would automatically move over to
that. And this was because the 286 chip was designed before the PC got popular. So,
you could either use the chip and emulate the old machine, and just have the speed.
Or, you could run it in the incompatible mode and get extra memory.
Now, there are a lot of crazy ideas about how to get around this limitation. IBM and
Intel fooled around with one that was kind of an emulation trick that involved using
the debug features of the chip. It didn't work for them, so they got me to help out. I
then basically proved that would never work. Then Gordon Letwin, who worked for
us, came up with another approach where you basically reset the chip all the time and
restart it in order to mix the two modes. So, it was kind of problematic. Clearly we
were running out of the 640K of memory. And like whenever any address base gets
pushed to its limit, you start doing what is called "bank switching" where instead of
the address telling you what to choose -- there is a state of the machine where it is
choosing different physical memory into the logical address base. And so the 286
looked like an opportunity. We did a version of our form of UNIX called XENIX for
the machine. Other people did high-end operating systems. But it just wasn't
powerful enough for that. So, the AT ended up being a high-speed DOS machine.
And a little bit of a distraction for the industry to figure out what to do with it. We
did run Windows on it, but in the final analysis, 286 machines were never really fast
enough for Windows. I mean they were nice machines but, during the time of the
286 machine, Windows really did not catch on.