I would like to share an idea that I believe could significantly improve the experience of working with AI coding agents, particularly Codex.
The idea: Continuous Conversation During Active Tasks
When working with coding agents, the current workflow often feels quite disconnected. You submit a prompt, the agent starts working, and you wait until it has finished or reached a point where you can interact with it again.
However, during that time, new ideas naturally come up. You might want to ask a question about the current implementation, change a detail, add a requirement or simply adjust the direction slightly.
I think it would be incredibly useful if users could have a continuous conversation with an agent while it is actively executing a task.
Instead of having to interrupt the process, submit another prompt and wait for the agent to continue, you could simply send additional messages at any point during execution.
For example, imagine asking Codex to build a new interface. While it is working, you realize that you would also like to change the color scheme. You simply tell it: “Could you also add green accents to the buttons, but keep the existing animations?”
The agent understands the additional instruction, incorporates it into its ongoing work and continues with the original task.
The same should apply to questions. You could ask why the agent chose a particular implementation, discuss an alternative or clarify something without having to stop the actual execution.
The key idea is that the conversation and the execution should happen simultaneously.
Ideally, the agent would maintain full awareness of its current progress, decisions and implementation state, allowing new instructions to be integrated intelligently without unnecessarily restarting or interrupting the task.
Of course, some changes might require adjusting the current plan or revisiting already completed work. The agent should be able to handle this intelligently and communicate when such adjustments are necessary.
I believe this would make working with AI coding agents feel much more natural. Software development is rarely a completely linear process. Ideas evolve, requirements change and new thoughts emerge while work is already underway.
Rather than treating every interaction as a separate prompt or task, the entire development process could become one continuous conversation between the user and the AI.
I think this could be a particularly valuable next step for Codex and AI-assisted software development in general.
Thank you for your time and for considering this suggestion. I really appreciate the work your team is doing and would love to see this concept explored in the future.
Best regards
Franz
