I'd go with Ant any day of the week.
It's not perfect; XML is very verbose and implementing any logic at all is almost impossible, but even the most junior engineer on the team can at least understand what the ant file is doing within a day.
Complex logic can be refactored using java and integrated in ant, if you so wish. Ant gives you all the power of java:)
Dependency resolution is difficult no matter what system you use. With ant, the best solutions seem to be either a lib directory in which all your jars are stored, or an internal web server from which the libraries are copied at build time.
I also have some experience with both Maven 1 and Maven 2. That experience left me with the feeling that Maven is awesome for hobby projects, but can have complications for software you need to maintain over time.
I see two important problems with Maven:
- Any dependency you import using Maven may change over time without you knowing it, resulting in weird problems.
- You import the licenses of not only the software you import directly using Maven, but also the licenses used by the libraries which are indirectly imported
In short, I dislike the fact that the build depends on the time it is started. That can be a true problem when producing a bugfix release a year after the production release.
These problems can of course be managed (maybe using a nexus proxy) but you need to consider them before rebuilding the build system. At my company we decided to use Ant for all new project and try to port maven 1 and 2 to ant whenever the occasion presents itself. It's just too difficult to keep it working.
My advice, if you and your team know how to deal with ant, try to refactor your ant file and don't jump on some other build tool. It just takes too much time to get it right; time you could spend making money and surviving as a company :)