In reality, all computer tasks require related programs to run and these programs could be considered agents. However, operations can generally be performed on a controlling machine that may use an agent, while its target is agentless in that it need not install or run new software related to the task itself. This capacity can save time required for managing agents on numerous target machines, especially in large enterprises. That said, even some software called agentless may use built-in services in an operating system, and they may require configuring. So the term agentless can be a bit misleading.
Agentless software generally requires the creation of a user account on the target machine or machines to facilitate access to the data on the account related to the desired operation. Agentless programs directly access the files, often remotely, via this user account. So the created profile must have the necessary access to these files and the software must store the login credentials to access the profile. This access may be facilitated through a number of different Internet communication standards (e.g., FTP, CIFS/SMB, TELNET, SSH).
Generally most home, SOHO, and SMB don’t need to narrow their choices between agent-based and agentless solutions. Most single computers and small networks have no need for agentless software. In enterprises, there may be considerations like tight security for trade secrets and the logistics of managing agents on thousands of machines. In the end, it comes down to deciding what works best for the particular company. Some may consider the risk of stored login credentials too great a risk even when encrypted.