When output
outruns review.

I began by writing a prompt, waiting for a result and deciding the next step. I later built my own voice agent. My workflow has since moved from a graphical dashboard and chat interface towards starting work through speech.

Speaking makes it easier to express an intention while staying with the question I am working through. The interface changes; the need to make the task and its boundaries understandable remains.

I now initiate parallel tasks by voice: research, editing and other knowledge work can move forward at the same time. I can start the next task while another continues.

This makes initiation feel less like the limiting step. Several streams of output can reach me before I have processed the first one.

The constraint shifts towards processing, reviewing, prioritising and combining the results. Starting more work does not automatically increase how much I can meaningfully assess.

Which outputs need immediate review? What should wait? Could clearer handovers or fewer concurrent tasks help? How should related results be combined? Those are design questions I am still working through.

Another challenge I see is recurring review cycles with AI and other people. Each round can produce another draft or document, while less attention goes to putting the result into practice and checking what it changes. Review and documentation matter when they clarify a decision or next step. How do I recognise when a result is ready to use, and when another round would mainly generate more text?