I think the interface is less important than the quality of the feedback loop.
For fast iteration, chat-style interfaces seem natural. But when working on large codebases, I still prefer something closer to an IDE/workflow, because context switching becomes expensive very quickly.
The interesting part of agentic engineering is no longer just “talking to the AI” — it’s how well the system understands the project context, constraints, previous decisions, and team workflow over time.
I’m curious to see how others are balancing speed and control here.
agreed - i like being able to choose and pick how i prefer to work, not locked into one interface, one workflow, one way of doing things. terminals might feel too "zoomed out" from the work sometimes.
Report
Less confortable with CLI... IDE mostly cuz it doesnt feel cramped and gives me a visual vantage point. Using CLI is a side kick that gets all the smaller tasks done without my monitoring.
Report
Depends on what you are doing, and how. If you need to add something small, ask for suggestion, check how a specific LLM behaves, so that you can gain some trust, so that you can teach it, having IDE would be really helpful.
A lot of us started with UI tools like Antigravity, Cursor.
Once things get advanced, the CLI works really good, it is faster, it does not lag. It has direct access to the console.
When I worked with Cursor and Antigravity, every time I wanted a analysis of the log, it saved the log in a file. Asked me for permission to save, asked for permission to read. And never deleted its left garbage of temporary log files. It polluted my repository, and created a risk for deletion of something important, while trying to clean the useless junk ;)
Report
IDE for me personally.
I still prefer seeing everything visually, especially when working on product design and mobile UI. But AI agents definitely make iteration much faster now.
Replies
Earth.fm
I think the interface is less important than the quality of the feedback loop.
For fast iteration, chat-style interfaces seem natural. But when working on large codebases, I still prefer something closer to an IDE/workflow, because context switching becomes expensive very quickly.
The interesting part of agentic engineering is no longer just “talking to the AI” — it’s how well the system understands the project context, constraints, previous decisions, and team workflow over time.
I’m curious to see how others are balancing speed and control here.
Kilo Code
exactly - you should get to choose how you work.
I feel the cli is much more interactive and can be used within vscode in any case. I prefer claude code but have also used codex in the same manner.
Kilo Code
agreed - i like being able to choose and pick how i prefer to work, not locked into one interface, one workflow, one way of doing things. terminals might feel too "zoomed out" from the work sometimes.
Less confortable with CLI... IDE mostly cuz it doesnt feel cramped and gives me a visual vantage point. Using CLI is a side kick that gets all the smaller tasks done without my monitoring.
Depends on what you are doing, and how. If you need to add something small, ask for suggestion, check how a specific LLM behaves, so that you can gain some trust, so that you can teach it, having IDE would be really helpful.
A lot of us started with UI tools like Antigravity, Cursor.
Once things get advanced, the CLI works really good, it is faster, it does not lag. It has direct access to the console.
When I worked with Cursor and Antigravity, every time I wanted a analysis of the log, it saved the log in a file. Asked me for permission to save, asked for permission to read. And never deleted its left garbage of temporary log files. It polluted my repository, and created a risk for deletion of something important, while trying to clean the useless junk ;)
IDE for me personally.
I still prefer seeing everything visually, especially when working on product design and mobile UI. But AI agents definitely make iteration much faster now.