Everyone is designing with words

The shift nobody’s talking about

I recently asked Cursor to change the text colour on my website to purple. It did something close to magenta, not quite what I wanted. So I tried again, this time more specifically: mauve, I said. Make it more deep purple, not lavender. That got it exactly right.

There are two differences to note.

1. I had a pretty precise idea of exactly what I wanted (and what I didn’t).
2. I was able to articulate it precisely (ie designing with words)

Amidst all the buzz about AI, there’s a quietly significant, more analogue, shift that is occurring across design and engineering practitioners, product managers, too. For the first time, the primary tool many of us reach for isn’t a component library or a Figma frame. It’s a sentence.

We are designing outputs, shaping interfaces, and writing systems into existence through natural language. A NYT article recently called this phenomenon a “talk fest” though this sounds quite casual for the nature of the language we’re dictating. The quality of that language input matters substantially. Content designers have often preached that we “design with words.” This feels only more universal in the new agentic world.

Content design—long the niche discipline, looped in late or not at all—turns out to have the exact competencies for the skill of the moment. Content designers have always practiced something that’s now relevant to everyone.

Clarity and precision are no longer nice-to-haves, they’re table stakes for prompt engineering.

Prompting is a communication problem

Prompting is, at its core, an applied communication challenge. The principles that create effective prompts could map almost directly onto those that govern good writing: know your audience, close the gap between what you mean and what you say, be accurate and precise.

Even brevity is still crucial: the longer you spend writing up your prompt or tweaking it because you didn’t get the expected output, the less efficient your tool usage becomes.

Content designers are almost always good communicators, not only for their end-user but their internal audience as well—often stakeholders they’re trying to convince, engineers who need specs, collaborators who they’ve requested feedback from.

They level-set their work with a “mode” in mind—draft, discovery, brainstorm, refinement, finessing, final draft. While it’s likely common enough or best practice for all product disciplines, it’s now more critical to share with our new agentic partners.

A prompt written in discovery mode—open, iterative, hypothesis-forming—calls for a fundamentally different structure than one written toward a known end state. Conflating the two is one of the more common failures in my own prompting; what do I actually want?!

Models optimize for what you appear to want. Vague inputs produce vague—yet overly confident—outputs, and the surface coherence of a model’s response will mask how far it is in reality from inferring what you actually needed.

While our toolbox may be undergoing immense change, our thinking may not need as large an overhaul. Rough drafts will still matter. Sketching, storyboarding, and back of the napkin drawings are still useful as we humans are not built to perfectly articulate our thoughts to an agent without some self-refinement.

These are tools we can still use to some extent, likely in quicker iteration, to get us closer to expressing our thoughts in a chat field, the exact shade of purple we can visualize but not describe. A rough sketch is worth 1,000 first prompt words.

Specificity is a key skill

Specificity also helps. Not just word choice (though mauve vs. purple is a real-life first-world problem), but structural specificity. Being clear about scope, constraints, the criteria by which a good response should be evaluated.

This goal specification—the degree to which a user can accurately express their internal representation of a task—is akin to the principles of, and likely the future of, the field of Human Computer Interaction. The closer the expression over chat is to the goal in mind, the shorter the feedback loop and fewer tweaks.

This has been the role of the UX writer (and many other communication disciplines) from day one: Explicitly convey the goal in order to reduce friction or drive the intended action. To explain something well, you must understand it first yourself.

The economist John List recently summed up exactly this underlying tension:

“Explaining why an answer is almost correct but subtly off requires exactly the critical thinking skills that created the knowledge in the first place.”

That’s what makes this moment pivotal for communications, and anyone who can communicate well. Thinking will become synonymous with communicating. The “doing” will increasingly happen through direction and articulation via that “thinking.”

Through knowing what we want well enough to recognize when we’re not getting it, and knowing how to course correct. The value won’t be in just prompting models, but recognizing and building on the standard for what the task should produce.
The most interesting prompting isn’t happening in single exchanges.

It’s happening in the work of building shared context with a model over time: refining system prompts, developing consistent interaction patterns, getting to a point where fewer corrections are needed because the model has been shaped well enough to be genuinely expressive of your thinking.

The work of building shared context

The goal for our discipline—for everyone now designing with words—will become a working knowledge of manipulating a system to increasingly reflect your exact intent back to you with fewer distortions. It won’t matter what you wrote vs. your AI tools, but the output still must have taste.

The ability to recognise subtle inaccuracies or losses in meaning—to feel that gap between nearly right and actually right—depends on having built that knowledge yourself and understanding what it looks like to convey it correctly.

This is a communication design challenge. It requires the same elements good communication has always required: understanding what you actually mean before you say it, building shared vocabulary, noticing when the exchange is working and when it’s drifting, and knowing your audience, now more than ever, especially, if it is an agent.

The people who can do this well, who can feel the difference between a model aligned with their intent and one producing fluent yet mediocre noise (and who can close that gap through their craft), are doing something that has always been hard.

Saying exactly what they mean to the right person at the right time. A challenge content designers (and all associated titles over the years) have been tasked with, hired for, and risen to the occasion of over the past two decades.

That skillset is difficult to truly automate; if anything, it’s more in demand right now.