Generally, with the regular in-IDE agents you have the ability to easily intervene, correct and live-check. Considering the high fail rate of agents (depending on software complexity of course), that's required if you want to get anything done and not be slowed down by it.

Otherwise you'd always have to context switch, consider which git state it's actually working from, etc. - rather than just letting the code directly before you change in your IDE.

It's significantly lower cognitive load and has a higher success rate, in my experience.

But, of course: Highly depends on the software being written and the general code infrastructure.

The problem that I see with background agents in general is that not following along and making it interactive adds compounding interest to the cost of editing, reviewing and understanding the agent's output since it's not something I've seen come together firsthand.

Agents also very rarely are truly hands-off: most of my usage is walking through a pre-determined set of steps and course-correcting along the way. In my experience, having it run out of sight leads to heavier editing since smaller realignments couldn't be applied along the way.

I like Junie because it is integrated with my favorite IDE and that counts for a lot although I wish it was better integrated and searched for things using the IDE’s database as opposed to Find-String.

I like having conversations with my agent, asking questions so I know how things work, asking it to ask me questions, etc. Personally for me one benefit of AI coding is better quality and better understanding. If it’s not clear how to do something with a certain library for instance I check it out of GitHub and point IntelliJ IDEA at it and ask Junie.

>> People are doubtful that the agent will be able to complete the task properly.

You answered your own question.

I do not trust an agent to give it unsupervised access to my systems.

If I had a completely local agent that was fully sandboxed and I would be willing to put data in the sandbox, give it a task, and come back later to see what it did.

I would not trust agents to run unsupervised with similar restrictions.

I haven't used them because I'm a big learn-by-doing guy that is constantly looking to expand or strengthen my skillset. Using a background coding agent takes all of the tinkering and debugging out of it, which is great if I just want results quickly, but completely counterintuitive if my goal is to become a better developer/engineer/architect/whatever.

they need too much hand holding still imho

i don't face any of these but sometimes i see agent gave me more than enough lines of code

i don't face any of these but sometimes ai agents give me more that enough lines of code