Coding agents like Claude Code and Codex dramatically speed up development, but a key bottleneck has emerged: aligning developer intent with agent output. When intent isn't aligned, agents implement features incorrectly, requiring multiple review cycles and wasting time. The author identifies three main bottlenecks: reviewing code, deciding what to work on, and aligning intent with coding agents.

To improve alignment, the author recommends using plan mode, where the agent does a read-only sweep and asks clarifying questions. This shifts the questioning burden to the agent, making communication more efficient. For smaller tasks, detailed prompts with rich context are sufficient.

The most effective technique is asking the model: "Is my understanding correct: [present understanding]". This is used 10-20 times daily. It allows the coding agent to verify or correct the developer's understanding, providing complete visibility into the work being done. This approach ensures both parties have aligned expectations before implementation begins.

The author emphasizes that protecting time is crucial now that we can build more in less time, making intent alignment optimization essential for engineering productivity.