As corroboration, my first assignment at Autonomy basically amounted to writing (on-site) a component that had been sold without the existence of a single line of code.
I think the obstacles for a startup wouldn't come from Autonomy's core techology---more the amount of stuff they've acquired around the edges. Pieces like KeyView (which is fairly good at reading just about any file format you care to mention) make the whole Autonomy package much more attractive than their search tech would be on its own. Plus, as you said, you have to be willing and able to compete with their sales force, and that seems like a soul-selling endeavor from the beginning.
Yes, the complete package is what they can offer big business. There are so many pieces to it now that they can turn it to a huge range of different uses.
The problem they have (well at least had when I was there a few years ago) is that nobody understands it all anymore. They've acquired so much technology and have such a high turn over rate of developers (barring a few extremely well rewarded key seniors) that its a constant uphill struggle to change anything or improve significantly, so developers have to just patch things up as best they can.
That burden will eventually catch up with them I would think, but I'm always surprised at how long a bad code-base can be kept alive.
I think the obstacles for a startup wouldn't come from Autonomy's core techology---more the amount of stuff they've acquired around the edges. Pieces like KeyView (which is fairly good at reading just about any file format you care to mention) make the whole Autonomy package much more attractive than their search tech would be on its own. Plus, as you said, you have to be willing and able to compete with their sales force, and that seems like a soul-selling endeavor from the beginning.