
AI and Human Creativity with John Voss of Descript
What should we keep doing ourselves when AI can handle more of the creative process? In this episode of the Torras AI Podcast, I speak with John Voss, Head of Product Design at Descript, about the relationship between creative tools, human judgment, and the work of making something worth sharing.
John's background includes work on the design system at Squarespace. His writing explores how people develop taste, how tools shape creative choices, and how design can give people more control over their work.
Watch the conversation
Start with the story, not the software
John draws a connection between building website tools at Squarespace and building video tools at Descript: people often arrive with something they want to communicate, rather than a desire to master a particular medium. The website or video is a way to tell that story. Designing for them starts with understanding what they are trying to accomplish.
That helps explain why deleting a sentence from a transcript can be such a powerful first experience. For someone making their first video, a single edit can turn an intimidating task into something they can imagine doing themselves. John still learns from watching that moment, even though text-based editing is a familiar part of his own working day.
Let AI find the material; keep the editorial judgment
Asked how he would use Descript's AI co-editor, Underlord, John describes his team's internal podcast. He begins by explaining the story he wants to tell and the intended length. He then asks the tool to highlight passages that support that narrative, including informative moments and humorous asides.
The first step is selection, rather than automatically cutting the recording. Highlights give him a manageable way into hours of material, while he supplies the direction and decides what belongs in the finished episode. The tool helps him review the footage; it does not determine what the conversation should mean.
When to chat, and when to grab a handle
Chat is useful when someone can describe an intention but does not know which controls will achieve it. John might ask for a vintage look, or for help finding relevant material across recordings. An AI assistant can translate that broad request into possible actions.
Precise adjustments are different. Setting a clip's first and last frames is often easier with a timeline handle than by describing timestamps. Fine-tuning contrast or black levels benefits from seeing the result immediately as a control moves. John argues for keeping these direct interactions available alongside chat: different tasks call for different ways of working.
Getting beyond the mediocre middle
AI can help a beginner go from making nothing to making something fairly polished. For an experienced creator, the benefit may be speed and capacity, while their own judgment continues to shape the result. John is interested in how tools can support distinctive work at both ends of that spectrum.
One answer is to resist making every workflow a race from prompt to finished output. Pausing between creative stages can give people opportunities to make choices. Another is to recognize the taste people already bring from other parts of their lives. A first-time video maker may still have a rich set of influences, references, and ideas. Tools could do more to help that person express those preferences.
A practical exercise for developing taste
In discussing his essay What is human taste?, John recalls advice from an art teacher: step away from the work. Spending too long looking at the same piece from the same position makes it harder to see the whole.
For video, that can mean watching on another screen, changing headphones, or viewing it with someone else. A different context can reveal an awkward cut or an overlong pause before publication. Comparing successive projects also helps: notice what improved, which effects you overuse, and which habits are becoming part of your voice. Taste develops through that repeated cycle of making, reviewing, and trying again.
Keep the effort that makes the work rewarding
We also discuss John's writing on AI and invasive design. His answer to which tasks should remain manual starts with personal satisfaction. He enjoys writing a first draft, then asking AI to help tighten it. Someone else may find that same task draining and want help at a different point.
The design challenge is to reduce chores without taking away the parts people value. John also distinguishes the difficulty of deciding what story to tell from the difficulty of operating software. Wrestling with a blank page can be worthwhile; being unable to work out how to start using the tool is a different problem.
Designing for different people improves the whole product
John gives accessibility as an example of how a specific need can lead to broader improvements. Keyboard navigation supports people who rely on it and can also make experienced editors faster. Clear structure and action labels help screen-reader users understand an interface; that clarity can also help agents operating software on someone's behalf.
The same reasoning applies to experience levels. Averaging beginners and experts into a single imaginary user can leave both poorly served. John describes a different ambition: consider both ends of the range, then find a coherent experience that supports them and the people in between.
Voice regeneration, consent, and audience expectations
We also discuss the responsibility that comes with tools that can generate speech in someone's voice. John distinguishes between editing an interview for length while preserving its meaning and generating words that a person never recorded. A news interview, a company update, and a fictional performance create different expectations.
For John, the speaker's consent is essential. Disclosure depends on the context and on what an audience would reasonably understand to be real. These are editorial decisions as well as product-design questions, and the availability of a feature does not settle whether it belongs in a particular piece of work.
What would better creativity look like?
Looking three years ahead, John would not judge success only by how polished AI-assisted work becomes. He would look for people continuing to create after the novelty wears off, established creators finding useful ways to expand their practice, and more people gaining the ability to tell their stories.
He would also look for experimentation. Trying a new medium or style can temporarily produce less polished work. That dip may be evidence of growth, rather than a failure of the tools. The hopeful outcome is a wider variety of work and more people developing a creative practice they want to sustain.
Follow John's work
Find John on LinkedIn and at jovo.design, where he shares his writing and other profiles. You can also explore Descript and his article on how AI creative tools can live up to their promise.





