True, they're not specifically licensing me to use their code in exchange, but I still think conceptually it's within my rights to offer that sort of exchange. I make a blanket offer to let you use my code, if in return you offer your own code under the same blanket offer. It's even more limited than that in that I'm not demanding you offer all your code that way, only the stuff that directly derives from and/or links with mine.
I don't see what's particularly dangerous about the AGPL, though. It's not that hard to comply with: if you build a webapp on some AGPL'd software, you need to AGPL the whole webapp. If you don't want to AGPL the whole webapp, you can just not build it on top of AGPL'd software. Pretty easy to avoid! And in any case the usual alternative to the AGPL is not releasing code at all, or releasing it under a "no commercial use" clause, which certainly doesn't give you more freedom. Compare Gitorious's code (AGPL) to GitHub's code (not released) for example. In both cases they're worried that if they MIT/BSD'd or even GPL'd their code, a rival SaaS could leapfrog them starting from their own code.
The example that really got to me with the AGPL was when the Opa programming language was being discussed on sites like HN a while back. IIRC, both the language and its standard library/runtime were AGPL'd. That would potentially mean that not just the compiler itself but any program you wrote in the language would automatically be AGPL'd as well.
As with all of these GPL-style licences, there's a lot more risk if you're not an all-GPL house, and quite modest risk if you're intending to GPL your code anyway. But the AGPL is different because it can affect not just developers building on the code for redistribution but also those who are just using it internally.
I can understand why they did that, in the context of network-enabled services, but nothing in the AGPL seems to restrict the implications to only that context. As the example above demonstrates, people do use it for other things with potentially profound implications that aren't obvious to anyone who knows about the much more common GPL and LGPL but doesn't realise why AGPL is fundamentally different. For that reason alone, I don't think the AGPL should have used a similar name.
I agree on the name, yeah, and I was also surprised by Opa. My assumption in infrastructure-code cases like that (not sure if actually true with Opa) is that they want to leave the door open to a dual-licensing solution, so if Opa ends up being a killer technology, the AGPL poison-pills it enough for commercial customers that they'll be able to sell paid licenses.
Not sure what a cleaner alternative would be in that case. Perhaps just state that explicitly? "You can use the free version of Opa to build open-source webapps [define appropriately], but need a commercial license if your app is not open-source."
I don't see what's particularly dangerous about the AGPL, though. It's not that hard to comply with: if you build a webapp on some AGPL'd software, you need to AGPL the whole webapp. If you don't want to AGPL the whole webapp, you can just not build it on top of AGPL'd software. Pretty easy to avoid! And in any case the usual alternative to the AGPL is not releasing code at all, or releasing it under a "no commercial use" clause, which certainly doesn't give you more freedom. Compare Gitorious's code (AGPL) to GitHub's code (not released) for example. In both cases they're worried that if they MIT/BSD'd or even GPL'd their code, a rival SaaS could leapfrog them starting from their own code.
[edited the 2nd paragraph above for brevity]