While I can easily argue the pro's and con's of both agent and agentless monitoring, in the end I almost always prefer agentless. I dislike agents for many of the reasons already mentioned here by others; difficult to manage all of the agents in large environments, typically result in a more costly solution, vendors typically lag behind in their support of new applications/OS's, need to be always tested before being rolled into production, etc.
One thing that I don't see mentioned here (apologies if I missed it) is utilizing API's. I think it's great when vendors like SolarWinds integrate with vendor API's such as the way SolarWinds does with VMWare. In this case you are leveraging a great solution that already exists versus "reinventing the wheel" by developing an agent or some other method to collect what is already made available by the API.
For a product like LEM with a strong security focus, having a "good" agent can be a big benefit for the secure transmission, compression, and offloading workload to the clients.
