Medium: The future of writing is better input, not AI writing

The most common framing in discussions of AI and writing puts the problem at the generation end: the output is generic, lacks voice, needs editing. Ryan Shrott’s piece on Medium challenges that framing by locating the real bottleneck earlier in the process. His argument is that the quality of AI-assisted writing is constrained not by how well AI generates but by how poorly humans capture their initial thinking before it erodes.

The piece focuses on what Shrott calls the capture stage — the moment when an idea is present and partially formed but has not yet been written down. He argues that most people impose order on their thinking too early, editing as they type in ways that flatten specificity and lose the texture of the original thought. By the time an idea reaches an AI tool, it has already been stripped of most of what made it worth having. Prompt engineering cannot recover what premature neatness has already discarded.

His proposed solution is voice-first capture: speaking rather than typing at the raw idea stage, using transcription to create a rough record that can later be cleaned and structured. The workflow he describes moves from speaking through transcript cleaning to AI-assisted organization and structure, with human refinement at the end. The key claim is that AI should work on thought, not manufacture it — and that thought, to be worth working on, needs to have been captured faithfully before it was tidied. Messy is not the problem at the capture stage; premature neatness is.

The argument has practical implications for how writers set up their work environment. It suggests that the investment worth making is not in prompt templates or output editing but in removing friction from the moment of raw capture — whether that means a voice tool, a dedicated low-pressure drafting environment, or a deliberate practice of speaking before typing. What goes into the pipeline determines what can come out.

Who this is useful for. Writers who find their AI-assisted drafts generic despite careful prompting, knowledge workers who dictate or journal but have not yet connected that practice to their AI writing tools, and content leads evaluating where to focus workflow improvement effort.