A Hacker News discussion reveals how AI coding assistants are disrupting developers' ability to enter flow states. Before agentic coding tools, one engineer prided themselves on sustained deep work, but now struggles with the stop-and-start rhythm of prompting slow models like Claude and waiting for responses.

The core issue is that chat-based interfaces create an incompatible workflow for programming. Developers describe the cycle as prompt, wait, wait, wait, check—more like managing an underling than collaborative coding. Flow state requires continuous concentration, but AI agents introduce interruptions that pull attention away from the code. Even faster models sacrifice intelligence, creating additional friction when fixes are needed.

Commenters propose several workarounds. One suggests comment-driven development—writing skeletons and comments manually, then letting agents fill implementation details. Others advocate for asynchronous workflows using file-based task managers instead of console prompts. The Design-Build-Test-Learn cycle emerges as a structured approach, where humans drive design while agents handle build and test phases.

The consensus points to a fundamental UI/UX problem in current AI coding tools. Generic chat interfaces fail to match how developers actually work, suggesting the field needs domain-specific interfaces rather than repurposed conversational patterns.