Agent or Agent-less for endpoint security, backup & F5 LB

For endpoint, agent is recommended as the HIPS need to conduct the scan, on demand and real time scan effectively. Agentless is not possible unless you are saying running as VM AV virtual appliance at hypervisor level.

For backup, in general, agent-based backup can be more effective as it is loaded in the OS stack it offers increased control and visibility of the host system that is not immediately available to an agent-less backup.  e.g. agent-less backups typically need to traverse the file system to determine changes for incremental/differential backups.  This traversal can take longer and be more complex than the agent-based backup which has access at the kernel level.

Also agent-based backups rely on local resources to pre-process and compress data before transmitting data across the network to the storage device (NAS/SAN, local USB drive, or remote host). I will say agent-less backups rely more on network resources to transmit application commands across the network as well as data between the target and storage device. Though both approaches will send data across the network taking up the bandwidth, the additional network traffic from agent-less backups could more likely to load the local network performance. Especially when your local network bandwidth is already trying to meet other Enterprise traffic.

But to put in context where most are in VM environment, the article shared the pro and xons and also agent assisted backup on top of the other two. It helps understanding and selecting the strategy.
http://www.veritas.com/community/blogs/agent-and-agentless-vm-backup-and-recovery-unraveling-myths

For F5, I have not heard there is agent besides the LB being able to be centrally manages if you have the F5 Enterprise Manager or Big-IQ that can save the config etc of each manage F5 device. This is preferred. Otherwise the manual backup via the console to snapshot the config, cert etc to store can still be done. Not so much of agent or agentless as for such application delivery controller, they already have API that can be called from external apps like those I shared. More of hybrid. So the focus should be on having to centrally manage of all your BIG-IP configuration files (UCS) which also serves as to maintain a central repository for BIG-IP configuration and information.

BIG-IQ Security uses snapshots to protect the working-configuration set of the Security module. Thus, at any time, you can back up, restore, and deploy the BIG-IQ working configuration to a specific configuration state, or deploy a specific set of working configuration edits back to a BIG-IP device. You can also compare one snapshot to another, or compare a snapshot to the BIG-IQ working configuration.

https://support.f5.com/kb/en-us/products/big-iq-security/manuals/product/bigiq-security-administration-4-4-0/11.html

May I assume that you are not familiar with the concepts of using an agent?
Agents "do" something on the remote machine - it could be anything from reading out status info or computer specs to modifying things like updating files or configurations. This could all be done from remote, agentless, too, all commands and procedures can be initiated fro a remote machine. The downside however is, that ports would need to be opened and a the machine controlling all this will be constantly instructing and checking remote machines - maybe it would even become a performance issue.

About your a-d, no answer can be given, it all depends on what particular software we look at and what we expect those agents to do. Normaly, an agent will not make machines instable, normally, it would not iinerfere with other apps.

Again normally, you would not always be given the choice, all endpoint security software that I know work with agents. Backups however can use built-in methods (wbadmin) depending on what you expect your backup to perform, this might be enough, so there's no agent. Some 3rd-party backup software use the task scheduler as control component.

I do as much as I can without extra software (and agents) using built-in methods, instead. Cheaper and mostly very reliable. But you will not always have a choice.